diff --git a/.dev_scripts/train_duo.sh b/.dev_scripts/train_duo.sh
new file mode 100755
index 0000000..b3f96d0
--- /dev/null
+++ b/.dev_scripts/train_duo.sh
@@ -0,0 +1,21 @@
+PARTITION=$1
+GPUS=${GPUS:-8}
+GPUS_PER_NODE=${GPUS_PER_NODE:-8}
+CPUS_PER_TASK=${CPUS_PER_TASK:-5}
+SRUN_ARGS=${SRUN_ARGS:-""}
+
+log_dir="work_dirs/slurm_logs"
+mkdir -p "$log_dir"
+
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-atss_r50_1x configs/detection/duo_dataset/atss_r50_fpn_1x_duo-coco.py work_dirs/duo/atss_r50_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/atss_r50_fpn_1x_duo-coco/eval_result > "$log_dir/duo_atss_r50_fpn_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-cascade_r50_1x configs/detection/duo_dataset/cascade-rcnn_r50_fpn_1x_duo-coco.py work_dirs/duo/cascade-rcnn_r50_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/cascade-rcnn_r50_fpn_1x_duo-coco/eval_result > "$log_dir/duo_cascade-rcnn_r50_fpn_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-faster_r50_1x configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_duo-coco.py work_dirs/duo/faster-rcnn_r50_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/faster-rcnn_r50_fpn_1x_duo-coco/eval_result > "$log_dir/duo_faster-rcnn_r50_fpn_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-faster_r101_1x configs/detection/duo_dataset/faster-rcnn_r101_fpn_1x_duo-coco.py work_dirs/duo/faster-rcnn_r101_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/faster-rcnn_r101_fpn_1x_duo-coco/eval_result > "$log_dir/duo_faster-rcnn_r101_fpn_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-faster_x101-32x4d_1x configs/detection/duo_dataset/faster-rcnn_x101-32x4d_fpn_1x_duo-coco.py work_dirs/duo/faster-rcnn_x101-32x4d_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/faster-rcnn_x101-32x4d_fpn_1x_duo-coco/eval_result > "$log_dir/duo_faster-rcnn_x101-32x4d_fpn_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-faster_x101-64x4d_1x configs/detection/duo_dataset/faster-rcnn_x101-64x4d_fpn_1x_duo-coco.py work_dirs/duo/faster-rcnn_x101-64x4d_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/faster-rcnn_x101-64x4d_fpn_1x_duo-coco/eval_result > "$log_dir/duo_faster-rcnn_x101-64x4d_fpn_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-fcos_r50_1x configs/detection/duo_dataset/fcos_r50-caffe_fpn_gn-head_1x_duo-coco.py work_dirs/duo/fcos_r50-caffe_fpn_gn-head_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/fcos_r50-caffe_fpn_gn-head_1x_duo-coco/eval_result > "$log_dir/duo_fcos_r50-caffe_fpn_gn-head_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-paa_r50_1x configs/detection/duo_dataset/paa_r50_fpn_1x_duo-coco.py work_dirs/duo/paa_r50_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/paa_r50_fpn_1x_duo-coco/eval_result > "$log_dir/duo_paa_r50_fpn_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-retinanet_r50_1x configs/detection/duo_dataset/retinanet_r50_fpn_1x_duo-coco.py work_dirs/duo/retinanet_r50_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/retinanet_r50_fpn_1x_duo-coco/eval_result > "$log_dir/duo_retinanet_r50_fpn_1x_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-ssd300 configs/detection/duo_dataset/ssd300_120e_duo-coco.py work_dirs/duo/ssd300_120e_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/ssd300_120e_duo-coco/eval_result > "$log_dir/duo_ssd300_120e_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-ssd512 configs/detection/duo_dataset/ssd512_120e_duo-coco.py work_dirs/duo/ssd512_120e_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/ssd512_120e_duo-coco/eval_result > "$log_dir/duo_ssd512_120e_duo-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION duo-tood_r50_1x configs/detection/duo_dataset/tood_r50_fpn_1x_duo-coco.py work_dirs/duo/tood_r50_fpn_1x_duo-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/duo/tood_r50_fpn_1x_duo-coco/eval_result > "$log_dir/duo_tood_r50_fpn_1x_duo-coco.log" &
diff --git a/.dev_scripts/train_rtts.sh b/.dev_scripts/train_rtts.sh
new file mode 100755
index 0000000..23677da
--- /dev/null
+++ b/.dev_scripts/train_rtts.sh
@@ -0,0 +1,16 @@
+PARTITION=$1
+GPUS=${GPUS:-8}
+GPUS_PER_NODE=${GPUS_PER_NODE:-8}
+CPUS_PER_TASK=${CPUS_PER_TASK:-5}
+SRUN_ARGS=${SRUN_ARGS:-""}
+
+log_dir="work_dirs/slurm_logs"
+mkdir -p "$log_dir"
+
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION rtts-atss_r50_1x configs/detection/rtts_dataset/atss_r50_fpn_1x_rtts-coco.py work_dirs/rtts/atss_r50_fpn_1x_rtts-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/rtts/atss_r50_fpn_1x_rtts-coco/eval_result > "$log_dir/rtts_atss_r50_fpn_1x_rtts-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION rtts-cascade_r50_1x configs/detection/rtts_dataset/cascade-rcnn_r50_fpn_1x_rtts-coco.py work_dirs/rtts/cascade-rcnn_r50_fpn_1x_rtts-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/rtts/cascade-rcnn_r50_fpn_1x_rtts-coco/eval_result > "$log_dir/rtts_cascade-rcnn_r50_fpn_1x_rtts-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION rtts-faster_r50_1x configs/detection/rtts_dataset/faster-rcnn_r50_fpn_1x_rtts-coco.py work_dirs/rtts/faster-rcnn_r50_fpn_1x_rtts-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/rtts/faster-rcnn_r50_fpn_1x_rtts-coco/eval_result > "$log_dir/rtts_faster-rcnn_r50_fpn_1x_rtts-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION rtts-fcos_r50_1x configs/detection/rtts_dataset/fcos_r50-caffe_fpn_gn-head_1x_rtts-coco.py work_dirs/rtts/fcos_r50-caffe_fpn_gn-head_1x_rtts-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/rtts/fcos_r50-caffe_fpn_gn-head_1x_rtts-coco/eval_result > "$log_dir/rtts_fcos_r50-caffe_fpn_gn-head_1x_rtts-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION rtts-paa_r50_1x configs/detection/rtts_dataset/paa_r50_fpn_1x_rtts-coco.py work_dirs/rtts/paa_r50_fpn_1x_rtts-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/rtts/paa_r50_fpn_1x_rtts-coco/eval_result > "$log_dir/rtts_paa_r50_fpn_1x_rtts-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION rtts-retinanet_r50_1x configs/detection/rtts_dataset/retinanet_r50_fpn_1x_rtts-coco.py work_dirs/rtts/retinanet_r50_fpn_1x_rtts-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/rtts/retinanet_r50_fpn_1x_rtts-coco/eval_result > "$log_dir/rtts_retinanet_r50_fpn_1x_rtts-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION rtts-tood_r50_1x configs/detection/rtts_dataset/tood_r50_fpn_1x_rtts-coco.py work_dirs/rtts/tood_r50_fpn_1x_rtts-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/rtts/tood_r50_fpn_1x_rtts-coco/eval_result > "$log_dir/rtts_tood_r50_fpn_1x_rtts-coco.log" &
diff --git a/.dev_scripts/train_uprc2020-train_all.sh b/.dev_scripts/train_uprc2020-train_all.sh
new file mode 100755
index 0000000..b2baa12
--- /dev/null
+++ b/.dev_scripts/train_uprc2020-train_all.sh
@@ -0,0 +1,21 @@
+PARTITION=$1
+GPUS=${GPUS:-8}
+GPUS_PER_NODE=${GPUS_PER_NODE:-8}
+CPUS_PER_TASK=${CPUS_PER_TASK:-5}
+SRUN_ARGS=${SRUN_ARGS:-""}
+
+log_dir="work_dirs/slurm_logs"
+mkdir -p "$log_dir"
+
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-atss_r50_1x configs/detection/urpc2020_dataset/train-all_test-A/atss_r50_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/atss_r50_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/atss_r50_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_atss_r50_fpn_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-cascade_r50_1x configs/detection/urpc2020_dataset/train-all_test-A/cascade-rcnn_r50_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/cascade-rcnn_r50_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/cascade-rcnn_r50_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_cascade-rcnn_r50_fpn_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-faster_r50_1x configs/detection/urpc2020_dataset/train-all_test-A/faster-rcnn_r50_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/faster-rcnn_r50_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/faster-rcnn_r50_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_faster-rcnn_r50_fpn_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-faster_r101_1x configs/detection/urpc2020_dataset/train-all_test-A/faster-rcnn_r101_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/faster-rcnn_r101_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/faster-rcnn_r101_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_faster-rcnn_r101_fpn_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-faster_x101-32x4d_1x configs/detection/urpc2020_dataset/train-all_test-A/faster-rcnn_x101-32x4d_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/faster-rcnn_x101-32x4d_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/faster-rcnn_x101-32x4d_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_faster-rcnn_x101-32x4d_fpn_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-faster_x101-64x4d_1x configs/detection/urpc2020_dataset/train-all_test-A/faster-rcnn_x101-64x4d_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/faster-rcnn_x101-64x4d_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/faster-rcnn_x101-64x4d_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_faster-rcnn_x101-64x4d_fpn_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-fcos_r50_1x configs/detection/urpc2020_dataset/train-all_test-A/fcos_r50-caffe_fpn_gn-head_1x_urpc-coco.py work_dirs/urpc2020-train-all/fcos_r50-caffe_fpn_gn-head_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/fcos_r50-caffe_fpn_gn-head_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_fcos_r50-caffe_fpn_gn-head_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-paa_r50_1x configs/detection/urpc2020_dataset/train-all_test-A/paa_r50_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/paa_r50_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/paa_r50_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_paa_r50_fpn_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-retinanet_r50_1x configs/detection/urpc2020_dataset/train-all_test-A/retinanet_r50_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/retinanet_r50_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/retinanet_r50_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_retinanet_r50_fpn_1x_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-ssd300 configs/detection/urpc2020_dataset/train-all_test-A/ssd300_120e_urpc-coco.py work_dirs/urpc2020-train-all/ssd300_120e_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/ssd300_120e_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_ssd300_120e_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-ssd512 configs/detection/urpc2020_dataset/train-all_test-A/ssd512_120e_urpc-coco.py work_dirs/urpc2020-train-all/ssd512_120e_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/ssd512_120e_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_ssd512_120e_urpc-coco.log" &
+SRUN_ARGS=${SRUN_ARGS} GPUS=$GPUS GPUS_PER_NODE=$GPUS_PER_NODE CPUS_PER_TASK=$CPUS_PRE_TASK ./tools/slurm_train.sh $PARTITION urpc2020-tood_r50_1x configs/detection/urpc2020_dataset/train-all_test-A/tood_r50_fpn_1x_urpc-coco.py work_dirs/urpc2020-train-all/tood_r50_fpn_1x_urpc-coco --cfg-options default_hooks.checkpoint.max_keep_ckpts=1 randomness.seed=0 val_evaluator.outfile_prefix=work_dirs/urpc2020-train-all/tood_r50_fpn_1x_urpc-coco/eval_result > "$log_dir/urpc2020-train-all_tood_r50_fpn_1x_urpc-coco.log" &
diff --git a/.gitignore b/.gitignore
index f323d04..70e4220 100644
--- a/.gitignore
+++ b/.gitignore
@@ -121,3 +121,6 @@ work_dirs/
*.pth
*.py~
*.sh~
+
+# ignore lark config
+configs/lark/
diff --git a/configs/detection/_base_/datasets/duo_coco_detection.py b/configs/detection/_base_/datasets/duo_coco_detection.py
new file mode 100644
index 0000000..d750084
--- /dev/null
+++ b/configs/detection/_base_/datasets/duo_coco_detection.py
@@ -0,0 +1,60 @@
+# dataset settings
+dataset_type = 'lqit.DUODataset' # `lqit` means the scope
+data_root = 'data/DUO/'
+
+backend_args = None
+
+train_pipeline = [
+ dict(type='LoadImageFromFile', backend_args=backend_args),
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(type='Resize', scale=(1333, 800), keep_ratio=True),
+ dict(type='RandomFlip', prob=0.5),
+ dict(type='PackDetInputs')
+]
+test_pipeline = [
+ dict(type='LoadImageFromFile', backend_args=backend_args),
+ dict(type='Resize', scale=(1333, 800), keep_ratio=True),
+ # If you don't have a gt annotation, delete the pipeline
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(
+ type='PackDetInputs',
+ meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+ 'scale_factor'))
+]
+train_dataloader = dict(
+ batch_size=2,
+ num_workers=2,
+ persistent_workers=True,
+ sampler=dict(type='DefaultSampler', shuffle=True),
+ batch_sampler=dict(type='AspectRatioBatchSampler'),
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ ann_file='annotations/instances_train.json',
+ data_prefix=dict(img='images/train/'),
+ filter_cfg=dict(filter_empty_gt=True, min_size=32),
+ pipeline=train_pipeline,
+ backend_args=backend_args))
+val_dataloader = dict(
+ batch_size=1,
+ num_workers=2,
+ persistent_workers=True,
+ drop_last=False,
+ sampler=dict(type='DefaultSampler', shuffle=False),
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ ann_file='annotations/instances_test.json',
+ data_prefix=dict(img='images/test/'),
+ test_mode=True,
+ pipeline=test_pipeline,
+ backend_args=backend_args))
+test_dataloader = val_dataloader
+
+val_evaluator = dict(
+ type='CocoMetric',
+ ann_file=data_root + 'annotations/instances_test.json',
+ metric='bbox',
+ format_only=False,
+ backend_args=backend_args)
+test_evaluator = val_evaluator
diff --git a/configs/detection/_base_/datasets/rtts_coco.py b/configs/detection/_base_/datasets/rtts_coco.py
index cbbb94e..f8b9ac3 100644
--- a/configs/detection/_base_/datasets/rtts_coco.py
+++ b/configs/detection/_base_/datasets/rtts_coco.py
@@ -1,6 +1,6 @@
# dataset settings
dataset_type = 'lqit.RTTSCocoDataset' # `lqit` means the scope
-data_root = 'data/RESIDE/'
+data_root = 'data/RESIDE/RTTS/'
backend_args = None
@@ -31,8 +31,8 @@
dataset=dict(
type=dataset_type,
data_root=data_root,
- ann_file='RTTS/annotations_json/rtts_train.json',
- data_prefix=dict(img='RTTS/'),
+ ann_file='annotations_json/train.json',
+ data_prefix=dict(img='JPEGImages/'),
filter_cfg=dict(filter_empty_gt=True, min_size=32),
pipeline=train_pipeline,
backend_args=backend_args))
@@ -45,8 +45,8 @@
dataset=dict(
type=dataset_type,
data_root=data_root,
- ann_file='RTTS/annotations_json/rtts_val.json',
- data_prefix=dict(img='RTTS/'),
+ ann_file='annotations_json/test.json',
+ data_prefix=dict(img='JPEGImages/'),
test_mode=True,
pipeline=test_pipeline,
backend_args=backend_args))
@@ -54,7 +54,7 @@
val_evaluator = dict(
type='CocoMetric',
- ann_file=data_root + 'RTTS/annotations_json/rtts_val.json',
+ ann_file=data_root + 'annotations_json/test.json',
metric='bbox',
format_only=False,
backend_args=backend_args)
diff --git a/configs/detection/_base_/datasets/urpc2020/urpc2020-test-A_coco_detection.py b/configs/detection/_base_/datasets/urpc2020/urpc2020-test-A_coco_detection.py
new file mode 100644
index 0000000..ef8593c
--- /dev/null
+++ b/configs/detection/_base_/datasets/urpc2020/urpc2020-test-A_coco_detection.py
@@ -0,0 +1,60 @@
+# dataset settings
+dataset_type = 'lqit.URPCCocoDataset' # `lqit` means the scope
+data_root = 'data/URPC2020/'
+
+backend_args = None
+
+train_pipeline = [
+ dict(type='LoadImageFromFile', backend_args=backend_args),
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(type='Resize', scale=(1333, 800), keep_ratio=True),
+ dict(type='RandomFlip', prob=0.5),
+ dict(type='PackDetInputs')
+]
+test_pipeline = [
+ dict(type='LoadImageFromFile', backend_args=backend_args),
+ dict(type='Resize', scale=(1333, 800), keep_ratio=True),
+ # If you don't have a gt annotation, delete the pipeline
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(
+ type='PackDetInputs',
+ meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+ 'scale_factor'))
+]
+train_dataloader = dict(
+ batch_size=2,
+ num_workers=2,
+ persistent_workers=True,
+ sampler=dict(type='DefaultSampler', shuffle=True),
+ batch_sampler=dict(type='AspectRatioBatchSampler'),
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ ann_file='annotations/train_all.json',
+ data_prefix=dict(img='train-image/'),
+ filter_cfg=dict(filter_empty_gt=True, min_size=32),
+ pipeline=train_pipeline,
+ backend_args=backend_args))
+val_dataloader = dict(
+ batch_size=1,
+ num_workers=2,
+ persistent_workers=True,
+ drop_last=False,
+ sampler=dict(type='DefaultSampler', shuffle=False),
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ ann_file='annotations/test-A.json',
+ data_prefix=dict(img='test-A-image/'),
+ test_mode=True,
+ pipeline=test_pipeline,
+ backend_args=backend_args))
+test_dataloader = val_dataloader
+
+val_evaluator = dict(
+ type='CocoMetric',
+ ann_file=data_root + 'annotations/test-A.json',
+ metric='bbox',
+ format_only=False,
+ backend_args=backend_args)
+test_evaluator = val_evaluator
diff --git a/configs/detection/_base_/datasets/urpc2020/urpc2020-test-B_coco_detection.py b/configs/detection/_base_/datasets/urpc2020/urpc2020-test-B_coco_detection.py
new file mode 100644
index 0000000..f5fbbd5
--- /dev/null
+++ b/configs/detection/_base_/datasets/urpc2020/urpc2020-test-B_coco_detection.py
@@ -0,0 +1,60 @@
+# dataset settings
+dataset_type = 'lqit.URPCCocoDataset' # `lqit` means the scope
+data_root = 'data/URPC2020/'
+
+backend_args = None
+
+train_pipeline = [
+ dict(type='LoadImageFromFile', backend_args=backend_args),
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(type='Resize', scale=(1333, 800), keep_ratio=True),
+ dict(type='RandomFlip', prob=0.5),
+ dict(type='PackDetInputs')
+]
+test_pipeline = [
+ dict(type='LoadImageFromFile', backend_args=backend_args),
+ dict(type='Resize', scale=(1333, 800), keep_ratio=True),
+ # If you don't have a gt annotation, delete the pipeline
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(
+ type='PackDetInputs',
+ meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+ 'scale_factor'))
+]
+train_dataloader = dict(
+ batch_size=2,
+ num_workers=2,
+ persistent_workers=True,
+ sampler=dict(type='DefaultSampler', shuffle=True),
+ batch_sampler=dict(type='AspectRatioBatchSampler'),
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ ann_file='annotations/train_all.json',
+ data_prefix=dict(img='train-image/'),
+ filter_cfg=dict(filter_empty_gt=True, min_size=32),
+ pipeline=train_pipeline,
+ backend_args=backend_args))
+val_dataloader = dict(
+ batch_size=1,
+ num_workers=2,
+ persistent_workers=True,
+ drop_last=False,
+ sampler=dict(type='DefaultSampler', shuffle=False),
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ ann_file='annotations/test-B.json',
+ data_prefix=dict(img='test-B-image/'),
+ test_mode=True,
+ pipeline=test_pipeline,
+ backend_args=backend_args))
+test_dataloader = val_dataloader
+
+val_evaluator = dict(
+ type='CocoMetric',
+ ann_file=data_root + 'annotations/test-B.json',
+ metric='bbox',
+ format_only=False,
+ backend_args=backend_args)
+test_evaluator = val_evaluator
diff --git a/configs/detection/_base_/datasets/urpc-2020_coco_detection.py b/configs/detection/_base_/datasets/urpc2020/urpc2020-validation_coco_detection.py
similarity index 86%
rename from configs/detection/_base_/datasets/urpc-2020_coco_detection.py
rename to configs/detection/_base_/datasets/urpc2020/urpc2020-validation_coco_detection.py
index 2363f59..3891837 100644
--- a/configs/detection/_base_/datasets/urpc-2020_coco_detection.py
+++ b/configs/detection/_base_/datasets/urpc2020/urpc2020-validation_coco_detection.py
@@ -1,6 +1,6 @@
# dataset settings
dataset_type = 'lqit.URPCCocoDataset' # `lqit` means the scope
-data_root = 'data/URPC/'
+data_root = 'data/URPC2020/'
backend_args = None
@@ -30,8 +30,8 @@
dataset=dict(
type=dataset_type,
data_root=data_root,
- ann_file='annotations_json/train.json',
- data_prefix=dict(img='JPEGImages/'),
+ ann_file='annotations/train.json',
+ data_prefix=dict(img='train-image/'),
filter_cfg=dict(filter_empty_gt=True, min_size=32),
pipeline=train_pipeline,
backend_args=backend_args))
@@ -44,8 +44,8 @@
dataset=dict(
type=dataset_type,
data_root=data_root,
- ann_file='annotations_json/val.json',
- data_prefix=dict(img='JPEGImages/'),
+ ann_file='annotations/val.json',
+ data_prefix=dict(img='train-image/'),
test_mode=True,
pipeline=test_pipeline,
backend_args=backend_args))
@@ -53,7 +53,7 @@
val_evaluator = dict(
type='CocoMetric',
- ann_file=data_root + 'annotations_json/val.json',
+ ann_file=data_root + 'annotations/val.json',
metric='bbox',
format_only=False,
backend_args=backend_args)
diff --git a/configs/detection/_base_/datasets/urpc-2020_xml_detection.py b/configs/detection/_base_/datasets/urpc2020/urpc2020-validation_xml_detection.py
similarity index 82%
rename from configs/detection/_base_/datasets/urpc-2020_xml_detection.py
rename to configs/detection/_base_/datasets/urpc2020/urpc2020-validation_xml_detection.py
index c7825ad..950c40f 100644
--- a/configs/detection/_base_/datasets/urpc-2020_xml_detection.py
+++ b/configs/detection/_base_/datasets/urpc2020/urpc2020-validation_xml_detection.py
@@ -30,10 +30,10 @@
dataset=dict(
type=dataset_type,
data_root=data_root,
- ann_file='ImageSets/train.txt',
- meta_file='ImageMetas/train-image-metas.pkl',
- img_subdir='JPEGImages',
- ann_subdir='annotations_xml',
+ ann_file='source_data/ImageSets/train.txt',
+ meta_file='source_data/ImageMetas/train-image-metas.pkl',
+ img_subdir='train-image',
+ ann_subdir='source_data/train-box',
data_prefix=dict(sub_data_root=''),
filter_cfg=dict(filter_empty_gt=True, min_size=32, bbox_min_size=32),
pipeline=train_pipeline,
@@ -47,10 +47,10 @@
dataset=dict(
type=dataset_type,
data_root=data_root,
- ann_file='ImageSets/val.txt',
- meta_file='ImageMetas/val-image-metas.pkl',
- img_subdir='JPEGImages',
- ann_subdir='annotations_xml',
+ ann_file='source_data/ImageSets/val.txt',
+ meta_file='source_data/ImageMetas/val-image-metas.pkl',
+ img_subdir='train-image',
+ ann_subdir='source_data/train-box',
data_prefix=dict(sub_data_root=''),
test_mode=True,
pipeline=test_pipeline,
diff --git a/configs/detection/_base_/default_runtime.py b/configs/detection/_base_/default_runtime.py
index 8e8c4ac..d2817fe 100644
--- a/configs/detection/_base_/default_runtime.py
+++ b/configs/detection/_base_/default_runtime.py
@@ -8,7 +8,6 @@
sampler_seed=dict(type='DistSamplerSeedHook'),
visualization=dict(type='DetVisualizationHook'))
-# set seed = 0 as default
randomness = dict(seed=None, deterministic=False)
env_cfg = dict(
diff --git a/configs/detection/duo_dataset/README.md b/configs/detection/duo_dataset/README.md
index e074880..589f424 100644
--- a/configs/detection/duo_dataset/README.md
+++ b/configs/detection/duo_dataset/README.md
@@ -1,7 +1,6 @@
# Detecting Underwater Objects
> [Detecting Underwater Objects](https://arxiv.org/abs/2106.05681)
-> [Underwater Robot Professional Contest 2020](https://www.heywhale.com/home/competition/5e535a612537a0002ca864ac/content/0)
diff --git a/configs/detection/duo_dataset/atss_r50_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/atss_r50_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..b7875c0
--- /dev/null
+++ b/configs/detection/duo_dataset/atss_r50_fpn_1x_duo-coco.py
@@ -0,0 +1,86 @@
+_base_ = [
+ '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='ATSS',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output',
+ num_outs=5),
+ bbox_head=dict(
+ type='ATSSHead',
+ num_classes=4,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ ratios=[1.0],
+ octave_base_scale=8,
+ scales_per_octave=1,
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='GIoULoss', loss_weight=2.0),
+ loss_centerness=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0)),
+ # training and testing settings
+ train_cfg=dict(
+ assigner=dict(type='ATSSAssigner', topk=9),
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.6),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='atss_r50_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/cascade-rcnn_r50_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/cascade-rcnn_r50_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..a843236
--- /dev/null
+++ b/configs/detection/duo_dataset/cascade-rcnn_r50_fpn_1x_duo-coco.py
@@ -0,0 +1,204 @@
+_base_ = [
+ '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='CascadeRCNN',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ scales=[8],
+ ratios=[0.5, 1.0, 2.0],
+ strides=[4, 8, 16, 32, 64]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0]),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
+ roi_head=dict(
+ type='CascadeRoIHead',
+ num_stages=3,
+ stage_loss_weights=[1, 0.5, 0.25],
+ bbox_roi_extractor=dict(
+ type='SingleRoIExtractor',
+ roi_layer=dict(type='RoIAlign', output_size=7, sampling_ratio=0),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=[
+ dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=4,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0,
+ loss_weight=1.0)),
+ dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=4,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.05, 0.05, 0.1, 0.1]),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0,
+ loss_weight=1.0)),
+ dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=4,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.033, 0.033, 0.067, 0.067]),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0))
+ ]),
+ # model training and testing settings
+ train_cfg=dict(
+ rpn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ match_low_quality=True,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=0,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_pre=2000,
+ max_per_img=2000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=[
+ dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False),
+ dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.6,
+ neg_iou_thr=0.6,
+ min_pos_iou=0.6,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False),
+ dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.7,
+ min_pos_iou=0.7,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False)
+ ]),
+ test_cfg=dict(
+ rpn=dict(
+ nms_pre=1000,
+ max_per_img=1000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100)))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='cascade-rcnn_r50_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_r101_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/faster-rcnn_r101_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..a0fd4aa
--- /dev/null
+++ b/configs/detection/duo_dataset/faster-rcnn_r101_fpn_1x_duo-coco.py
@@ -0,0 +1,21 @@
+_base_ = 'faster-rcnn_r50_fpn_1x_duo-coco.py'
+
+model = dict(
+ backbone=dict(
+ depth=101,
+ init_cfg=dict(type='Pretrained',
+ checkpoint='torchvision://resnet101')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='faster-rcnn_r101_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_r101_fpn_2x_duo-coco.py b/configs/detection/duo_dataset/faster-rcnn_r101_fpn_2x_duo-coco.py
new file mode 100644
index 0000000..9a051f9
--- /dev/null
+++ b/configs/detection/duo_dataset/faster-rcnn_r101_fpn_2x_duo-coco.py
@@ -0,0 +1,21 @@
+_base_ = 'faster-rcnn_r50_fpn_2x_duo-coco.py'
+
+model = dict(
+ backbone=dict(
+ depth=101,
+ init_cfg=dict(type='Pretrained',
+ checkpoint='torchvision://resnet101')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='faster-rcnn_r101_fpn_2x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..de11f4e
--- /dev/null
+++ b/configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_duo-coco.py
@@ -0,0 +1,133 @@
+_base_ = [
+ '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='FasterRCNN',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ scales=[8],
+ ratios=[0.5, 1.0, 2.0],
+ strides=[4, 8, 16, 32, 64]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0]),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
+ roi_head=dict(
+ type='StandardRoIHead',
+ bbox_roi_extractor=dict(
+ type='SingleRoIExtractor',
+ roi_layer=dict(type='RoIAlign', output_size=7, sampling_ratio=0),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=4,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='L1Loss', loss_weight=1.0))),
+ # model training and testing settings
+ train_cfg=dict(
+ rpn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ match_low_quality=True,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_pre=2000,
+ max_per_img=1000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False)),
+ test_cfg=dict(
+ rpn=dict(
+ nms_pre=1000,
+ max_per_img=1000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100)
+ # soft-nms is also supported for rcnn testing
+ # e.g., nms=dict(type='soft_nms', iou_threshold=0.5, min_score=0.05)
+ ))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='faster-rcnn_r50_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_urpc-xml.py b/configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_urpc-xml.py
deleted file mode 100644
index 426038c..0000000
--- a/configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_urpc-xml.py
+++ /dev/null
@@ -1,7 +0,0 @@
-_base_ = [
- '../_base_/models/faster-rcnn_r50_fpn.py',
- '../_base_/datasets/urpc-2020_xml_detection.py',
- '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
-]
-
-model = dict(roi_head=dict(bbox_head=dict(num_classes=4)))
diff --git a/configs/detection/duo_dataset/faster-rcnn_r50_fpn_2x_duo-coco.py b/configs/detection/duo_dataset/faster-rcnn_r50_fpn_2x_duo-coco.py
new file mode 100644
index 0000000..6a7135f
--- /dev/null
+++ b/configs/detection/duo_dataset/faster-rcnn_r50_fpn_2x_duo-coco.py
@@ -0,0 +1,133 @@
+_base_ = [
+ '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_2x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='FasterRCNN',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ scales=[8],
+ ratios=[0.5, 1.0, 2.0],
+ strides=[4, 8, 16, 32, 64]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0]),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
+ roi_head=dict(
+ type='StandardRoIHead',
+ bbox_roi_extractor=dict(
+ type='SingleRoIExtractor',
+ roi_layer=dict(type='RoIAlign', output_size=7, sampling_ratio=0),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=4,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='L1Loss', loss_weight=1.0))),
+ # model training and testing settings
+ train_cfg=dict(
+ rpn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ match_low_quality=True,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_pre=2000,
+ max_per_img=1000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False)),
+ test_cfg=dict(
+ rpn=dict(
+ nms_pre=1000,
+ max_per_img=1000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100)
+ # soft-nms is also supported for rcnn testing
+ # e.g., nms=dict(type='soft_nms', iou_threshold=0.5, min_score=0.05)
+ ))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='faster-rcnn_r50_fpn_2x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_x101-32x4d_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/faster-rcnn_x101-32x4d_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..c8abe42
--- /dev/null
+++ b/configs/detection/duo_dataset/faster-rcnn_x101-32x4d_fpn_1x_duo-coco.py
@@ -0,0 +1,29 @@
+_base_ = 'faster-rcnn_r50_fpn_1x_duo-coco.py'
+
+model = dict(
+ backbone=dict(
+ type='ResNeXt',
+ depth=101,
+ groups=32,
+ base_width=4,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ style='pytorch',
+ init_cfg=dict(
+ type='Pretrained', checkpoint='open-mmlab://resnext101_32x4d')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='faster-rcnn_x101-32x4d_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_x101-32x4d_fpn_2x_duo-coco.py b/configs/detection/duo_dataset/faster-rcnn_x101-32x4d_fpn_2x_duo-coco.py
new file mode 100644
index 0000000..57419be
--- /dev/null
+++ b/configs/detection/duo_dataset/faster-rcnn_x101-32x4d_fpn_2x_duo-coco.py
@@ -0,0 +1,29 @@
+_base_ = 'faster-rcnn_r50_fpn_2x_duo-coco.py'
+
+model = dict(
+ backbone=dict(
+ type='ResNeXt',
+ depth=101,
+ groups=32,
+ base_width=4,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ style='pytorch',
+ init_cfg=dict(
+ type='Pretrained', checkpoint='open-mmlab://resnext101_32x4d')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='faster-rcnn_x101-32x4d_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_x101-64x4d_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/faster-rcnn_x101-64x4d_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..e9e9760
--- /dev/null
+++ b/configs/detection/duo_dataset/faster-rcnn_x101-64x4d_fpn_1x_duo-coco.py
@@ -0,0 +1,29 @@
+_base_ = 'faster-rcnn_r50_fpn_1x_duo-coco.py'
+
+model = dict(
+ backbone=dict(
+ type='ResNeXt',
+ depth=101,
+ groups=64,
+ base_width=4,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ style='pytorch',
+ init_cfg=dict(
+ type='Pretrained', checkpoint='open-mmlab://resnext101_64x4d')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='faster-rcnn_x101-64x4d_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_x101-64x4d_fpn_2x_duo-coco.py b/configs/detection/duo_dataset/faster-rcnn_x101-64x4d_fpn_2x_duo-coco.py
new file mode 100644
index 0000000..1e9de74
--- /dev/null
+++ b/configs/detection/duo_dataset/faster-rcnn_x101-64x4d_fpn_2x_duo-coco.py
@@ -0,0 +1,29 @@
+_base_ = 'faster-rcnn_r50_fpn_2x_duo-coco.py'
+
+model = dict(
+ backbone=dict(
+ type='ResNeXt',
+ depth=101,
+ groups=64,
+ base_width=4,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ style='pytorch',
+ init_cfg=dict(
+ type='Pretrained', checkpoint='open-mmlab://resnext101_64x4d')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='faster-rcnn_x101-64x4d_fpn_2x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/fcos_r50-caffe_fpn_gn-head_1x_duo-coco.py b/configs/detection/duo_dataset/fcos_r50-caffe_fpn_gn-head_1x_duo-coco.py
new file mode 100644
index 0000000..248bfea
--- /dev/null
+++ b/configs/detection/duo_dataset/fcos_r50-caffe_fpn_gn-head_1x_duo-coco.py
@@ -0,0 +1,91 @@
+_base_ = [
+ '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='FCOS',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[102.9801, 115.9465, 122.7717],
+ std=[1.0, 1.0, 1.0],
+ bgr_to_rgb=False,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=False),
+ norm_eval=True,
+ style='caffe',
+ init_cfg=dict(
+ type='Pretrained',
+ checkpoint='open-mmlab://detectron/resnet50_caffe')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output', # use P5
+ num_outs=5,
+ relu_before_extra_convs=True),
+ bbox_head=dict(
+ type='FCOSHead',
+ num_classes=4,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ strides=[8, 16, 32, 64, 128],
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='IoULoss', loss_weight=1.0),
+ loss_centerness=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0)),
+ # testing settings
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100))
+
+# learning rate
+param_scheduler = [
+ dict(
+ type='LinearLR', start_factor=0.001, by_epoch=False, begin=0,
+ end=1000),
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=12,
+ by_epoch=True,
+ milestones=[8, 11],
+ gamma=0.1)
+]
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(lr=0.01),
+ paramwise_cfg=dict(bias_lr_mult=2., bias_decay_mult=0.),
+ clip_grad=dict(max_norm=35, norm_type=2)) # loss may NaN without clip_grad
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='fcos_r50-caffe_fpn_gn-head_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/paa_r50_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/paa_r50_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..5f542cd
--- /dev/null
+++ b/configs/detection/duo_dataset/paa_r50_fpn_1x_duo-coco.py
@@ -0,0 +1,107 @@
+_base_ = [
+ '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='PAA',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output',
+ num_outs=5),
+ bbox_head=dict(
+ type='PAAHead',
+ reg_decoded_bbox=True,
+ score_voting=True,
+ topk=9,
+ num_classes=5,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ ratios=[1.0],
+ octave_base_scale=8,
+ scales_per_octave=1,
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='GIoULoss', loss_weight=1.3),
+ loss_centerness=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=0.5)),
+ # training and testing settings
+ train_cfg=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.1,
+ neg_iou_thr=0.1,
+ min_pos_iou=0,
+ ignore_iof_thr=-1),
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.6),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# learning rate
+param_scheduler = [
+ dict(
+ type='LinearLR', start_factor=0.001, by_epoch=False, begin=0, end=500),
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=12,
+ by_epoch=True,
+ milestones=[8, 11],
+ gamma=0.1)
+]
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='paa_r50_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/retinanet_r50_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/retinanet_r50_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..7c1d5f6
--- /dev/null
+++ b/configs/detection/duo_dataset/retinanet_r50_fpn_1x_duo-coco.py
@@ -0,0 +1,105 @@
+_base_ = [
+ '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='RetinaNet',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_input',
+ num_outs=5),
+ bbox_head=dict(
+ type='RetinaHead',
+ num_classes=4,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ octave_base_scale=4,
+ scales_per_octave=3,
+ ratios=[0.5, 1.0, 2.0],
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0]),
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
+ # model training and testing settings
+ train_cfg=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.4,
+ min_pos_iou=0,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='PseudoSampler'), # Focal loss should use PseudoSampler
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001),
+ clip_grad=dict(max_norm=35, norm_type=2)) # loss may NaN without clip_grad
+
+param_scheduler = [
+ dict(
+ type='LinearLR', start_factor=0.001, by_epoch=False, begin=0,
+ end=1000),
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=12,
+ by_epoch=True,
+ milestones=[8, 11],
+ gamma=0.1)
+]
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='retinanet_r50_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/ssd300_120e_duo-coco.py b/configs/detection/duo_dataset/ssd300_120e_duo-coco.py
new file mode 100644
index 0000000..0a0a857
--- /dev/null
+++ b/configs/detection/duo_dataset/ssd300_120e_duo-coco.py
@@ -0,0 +1,101 @@
+_base_ = [
+ '../_base_/models/ssd300.py', '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_2x.py', '../_base_/default_runtime.py'
+]
+# model settings
+model = dict(bbox_head=dict(num_classes=4))
+
+# dataset settings
+input_size = 300
+train_pipeline = [
+ dict(type='LoadImageFromFile'),
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(
+ type='Expand',
+ mean={{_base_.model.data_preprocessor.mean}},
+ to_rgb={{_base_.model.data_preprocessor.bgr_to_rgb}},
+ ratio_range=(1, 4)),
+ dict(
+ type='MinIoURandomCrop',
+ min_ious=(0.1, 0.3, 0.5, 0.7, 0.9),
+ min_crop_size=0.3),
+ dict(type='Resize', scale=(input_size, input_size), keep_ratio=False),
+ dict(type='RandomFlip', prob=0.5),
+ dict(
+ type='PhotoMetricDistortion',
+ brightness_delta=32,
+ contrast_range=(0.5, 1.5),
+ saturation_range=(0.5, 1.5),
+ hue_delta=18),
+ dict(type='PackDetInputs')
+]
+test_pipeline = [
+ dict(type='LoadImageFromFile'),
+ dict(type='Resize', scale=(input_size, input_size), keep_ratio=False),
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(
+ type='PackDetInputs',
+ meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+ 'scale_factor'))
+]
+train_dataloader = dict(
+ batch_size=8,
+ num_workers=2,
+ batch_sampler=None,
+ dataset=dict(
+ _delete_=True,
+ type='RepeatDataset',
+ times=5,
+ dataset=dict(
+ type={{_base_.dataset_type}},
+ data_root={{_base_.data_root}},
+ ann_file='annotations/instances_train.json',
+ data_prefix=dict(img='images/train/'),
+ filter_cfg=dict(filter_empty_gt=True, min_size=32),
+ pipeline=train_pipeline)))
+val_dataloader = dict(batch_size=8, dataset=dict(pipeline=test_pipeline))
+test_dataloader = val_dataloader
+
+# optimizer
+optim_wrapper = dict(
+ type='OptimWrapper',
+ optimizer=dict(type='SGD', lr=2e-3, momentum=0.9, weight_decay=5e-4),
+ clip_grad=dict(max_norm=35, norm_type=2)) # loss may NaN without clip_grad
+
+custom_hooks = [
+ dict(type='NumClassCheckHook'),
+ dict(type='CheckInvalidLossHook', interval=50, priority='VERY_LOW')
+]
+
+# learning rate
+param_scheduler = [
+ dict(
+ type='LinearLR', start_factor=0.001, by_epoch=False, begin=0,
+ end=1000),
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=24,
+ by_epoch=True,
+ milestones=[16, 22],
+ gamma=0.1)
+]
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# USER SHOULD NOT CHANGE ITS VALUES.
+# base_batch_size = (8 GPUs) x (8 samples per GPU)
+auto_scale_lr = dict(base_batch_size=64)
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='ssd300_120e__duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/ssd512_120e_duo-coco.py b/configs/detection/duo_dataset/ssd512_120e_duo-coco.py
new file mode 100644
index 0000000..921cedc
--- /dev/null
+++ b/configs/detection/duo_dataset/ssd512_120e_duo-coco.py
@@ -0,0 +1,74 @@
+_base_ = 'ssd300_120e_duo-coco.py'
+
+# model settings
+input_size = 512
+model = dict(
+ neck=dict(
+ out_channels=(512, 1024, 512, 256, 256, 256, 256),
+ level_strides=(2, 2, 2, 2, 1),
+ level_paddings=(1, 1, 1, 1, 1),
+ last_kernel_size=4),
+ bbox_head=dict(
+ in_channels=(512, 1024, 512, 256, 256, 256, 256),
+ anchor_generator=dict(
+ type='SSDAnchorGenerator',
+ scale_major=False,
+ input_size=input_size,
+ basesize_ratio_range=(0.1, 0.9),
+ strides=[8, 16, 32, 64, 128, 256, 512],
+ ratios=[[2], [2, 3], [2, 3], [2, 3], [2, 3], [2], [2]])))
+
+# dataset settings
+train_pipeline = [
+ dict(type='LoadImageFromFile'),
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(
+ type='Expand',
+ mean={{_base_.model.data_preprocessor.mean}},
+ to_rgb={{_base_.model.data_preprocessor.bgr_to_rgb}},
+ ratio_range=(1, 4)),
+ dict(
+ type='MinIoURandomCrop',
+ min_ious=(0.1, 0.3, 0.5, 0.7, 0.9),
+ min_crop_size=0.3),
+ dict(type='Resize', scale=(input_size, input_size), keep_ratio=False),
+ dict(type='RandomFlip', prob=0.5),
+ dict(
+ type='PhotoMetricDistortion',
+ brightness_delta=32,
+ contrast_range=(0.5, 1.5),
+ saturation_range=(0.5, 1.5),
+ hue_delta=18),
+ dict(type='PackDetInputs')
+]
+test_pipeline = [
+ dict(type='LoadImageFromFile'),
+ dict(type='Resize', scale=(input_size, input_size), keep_ratio=False),
+ dict(type='LoadAnnotations', with_bbox=True),
+ dict(
+ type='PackDetInputs',
+ meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+ 'scale_factor'))
+]
+train_dataloader = dict(dataset=dict(dataset=dict(pipeline=train_pipeline)))
+val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
+test_dataloader = val_dataloader
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# USER SHOULD NOT CHANGE ITS VALUES.
+# base_batch_size = (8 GPUs) x (8 samples per GPU)
+auto_scale_lr = dict(base_batch_size=64)
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='ssd512_120e_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/tood_r50_fpn_1x_duo-coco.py b/configs/detection/duo_dataset/tood_r50_fpn_1x_duo-coco.py
new file mode 100644
index 0000000..d105470
--- /dev/null
+++ b/configs/detection/duo_dataset/tood_r50_fpn_1x_duo-coco.py
@@ -0,0 +1,95 @@
+_base_ = [
+ '../_base_/datasets/duo_coco_detection.py',
+ '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='TOOD',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output',
+ num_outs=5),
+ bbox_head=dict(
+ type='TOODHead',
+ num_classes=4,
+ in_channels=256,
+ stacked_convs=6,
+ feat_channels=256,
+ anchor_type='anchor_free',
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ ratios=[1.0],
+ octave_base_scale=8,
+ scales_per_octave=1,
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ initial_loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ activated=True, # use probability instead of logit as input
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_cls=dict(
+ type='QualityFocalLoss',
+ use_sigmoid=True,
+ activated=True, # use probability instead of logit as input
+ beta=2.0,
+ loss_weight=1.0),
+ loss_bbox=dict(type='GIoULoss', loss_weight=2.0)),
+ train_cfg=dict(
+ initial_epoch=4,
+ initial_assigner=dict(type='ATSSAssigner', topk=9),
+ assigner=dict(type='TaskAlignedAssigner', topk=13),
+ alpha=1,
+ beta=6,
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.6),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='DUO_detection',
+# name='tood_r50_fpn_1x_duo',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/lqit/edit/hooks/__init__.py b/configs/detection/rtts_dataset/README.md
similarity index 100%
rename from lqit/edit/hooks/__init__.py
rename to configs/detection/rtts_dataset/README.md
diff --git a/configs/detection/rtts_dataset/atss_r50_fpn_1x_rtts-coco.py b/configs/detection/rtts_dataset/atss_r50_fpn_1x_rtts-coco.py
new file mode 100644
index 0000000..a274ead
--- /dev/null
+++ b/configs/detection/rtts_dataset/atss_r50_fpn_1x_rtts-coco.py
@@ -0,0 +1,86 @@
+_base_ = [
+ '../_base_/datasets/rtts_coco.py', '../_base_/schedules/schedule_1x.py',
+ '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='ATSS',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output',
+ num_outs=5),
+ bbox_head=dict(
+ type='ATSSHead',
+ num_classes=5,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ ratios=[1.0],
+ octave_base_scale=8,
+ scales_per_octave=1,
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='GIoULoss', loss_weight=2.0),
+ loss_centerness=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0)),
+ # training and testing settings
+ train_cfg=dict(
+ assigner=dict(type='ATSSAssigner', topk=9),
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.6),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='rtts_detection',
+# name='atss_r50_fpn_1x_rtts',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/rtts_dataset/cascade-rcnn_r50_fpn_1x_rtts-coco.py b/configs/detection/rtts_dataset/cascade-rcnn_r50_fpn_1x_rtts-coco.py
new file mode 100644
index 0000000..b575435
--- /dev/null
+++ b/configs/detection/rtts_dataset/cascade-rcnn_r50_fpn_1x_rtts-coco.py
@@ -0,0 +1,204 @@
+_base_ = [
+ '../_base_/datasets/rtts_coco.py', '../_base_/schedules/schedule_1x.py',
+ '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='CascadeRCNN',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ scales=[8],
+ ratios=[0.5, 1.0, 2.0],
+ strides=[4, 8, 16, 32, 64]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0]),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
+ roi_head=dict(
+ type='CascadeRoIHead',
+ num_stages=3,
+ stage_loss_weights=[1, 0.5, 0.25],
+ bbox_roi_extractor=dict(
+ type='SingleRoIExtractor',
+ roi_layer=dict(type='RoIAlign', output_size=7, sampling_ratio=0),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=[
+ dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=5,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0,
+ loss_weight=1.0)),
+ dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=5,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.05, 0.05, 0.1, 0.1]),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0,
+ loss_weight=1.0)),
+ dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=5,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.033, 0.033, 0.067, 0.067]),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0))
+ ]),
+ # model training and testing settings
+ train_cfg=dict(
+ rpn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ match_low_quality=True,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=0,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_pre=2000,
+ max_per_img=2000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=[
+ dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False),
+ dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.6,
+ neg_iou_thr=0.6,
+ min_pos_iou=0.6,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False),
+ dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.7,
+ min_pos_iou=0.7,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False)
+ ]),
+ test_cfg=dict(
+ rpn=dict(
+ nms_pre=1000,
+ max_per_img=1000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100)))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='rtts_detection',
+# name='cascade-rcnn_r50_fpn_1x_rtts',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/rtts_dataset/faster-rcnn_r50_fpn_1x_rtts-coco.py b/configs/detection/rtts_dataset/faster-rcnn_r50_fpn_1x_rtts-coco.py
new file mode 100644
index 0000000..02f15e5
--- /dev/null
+++ b/configs/detection/rtts_dataset/faster-rcnn_r50_fpn_1x_rtts-coco.py
@@ -0,0 +1,133 @@
+_base_ = [
+ '../_base_/datasets/rtts_coco.py', '../_base_/schedules/schedule_1x.py',
+ '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='FasterRCNN',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ scales=[8],
+ ratios=[0.5, 1.0, 2.0],
+ strides=[4, 8, 16, 32, 64]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0]),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
+ roi_head=dict(
+ type='StandardRoIHead',
+ bbox_roi_extractor=dict(
+ type='SingleRoIExtractor',
+ roi_layer=dict(type='RoIAlign', output_size=7, sampling_ratio=0),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=dict(
+ type='Shared2FCBBoxHead',
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=5,
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='L1Loss', loss_weight=1.0))),
+ # model training and testing settings
+ train_cfg=dict(
+ rpn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ match_low_quality=True,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_pre=2000,
+ max_per_img=1000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ match_low_quality=False,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ pos_weight=-1,
+ debug=False)),
+ test_cfg=dict(
+ rpn=dict(
+ nms_pre=1000,
+ max_per_img=1000,
+ nms=dict(type='nms', iou_threshold=0.7),
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100)
+ # soft-nms is also supported for rcnn testing
+ # e.g., nms=dict(type='soft_nms', iou_threshold=0.5, min_score=0.05)
+ ))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='rtts_detection',
+# name='faster-rcnn_r50_fpn_1x_rtts',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/rtts_dataset/fcos_r50-caffe_fpn_gn-head_1x_rtts-coco.py b/configs/detection/rtts_dataset/fcos_r50-caffe_fpn_gn-head_1x_rtts-coco.py
new file mode 100644
index 0000000..b57a75a
--- /dev/null
+++ b/configs/detection/rtts_dataset/fcos_r50-caffe_fpn_gn-head_1x_rtts-coco.py
@@ -0,0 +1,91 @@
+_base_ = [
+ '../_base_/datasets/rtts_coco.py', '../_base_/schedules/schedule_1x.py',
+ '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='FCOS',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[102.9801, 115.9465, 122.7717],
+ std=[1.0, 1.0, 1.0],
+ bgr_to_rgb=False,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=False),
+ norm_eval=True,
+ style='caffe',
+ init_cfg=dict(
+ type='Pretrained',
+ checkpoint='open-mmlab://detectron/resnet50_caffe')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output', # use P5
+ num_outs=5,
+ relu_before_extra_convs=True),
+ bbox_head=dict(
+ type='FCOSHead',
+ num_classes=5,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ strides=[8, 16, 32, 64, 128],
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='IoULoss', loss_weight=1.0),
+ loss_centerness=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0)),
+ # testing settings
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100))
+
+# learning rate
+param_scheduler = [
+ dict(
+ type='LinearLR', start_factor=0.001, by_epoch=False, begin=0,
+ end=1000),
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=12,
+ by_epoch=True,
+ milestones=[8, 11],
+ gamma=0.1)
+]
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(lr=0.01),
+ paramwise_cfg=dict(bias_lr_mult=2., bias_decay_mult=0.),
+ clip_grad=dict(max_norm=35, norm_type=2)) # loss may NaN without clip_grad
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='rtts_detection',
+# name='fcos_r50-caffe_fpn_gn-head_1x_rtts',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/rtts_dataset/paa_r50_fpn_1x_rtts-coco.py b/configs/detection/rtts_dataset/paa_r50_fpn_1x_rtts-coco.py
new file mode 100644
index 0000000..ae5f7eb
--- /dev/null
+++ b/configs/detection/rtts_dataset/paa_r50_fpn_1x_rtts-coco.py
@@ -0,0 +1,107 @@
+_base_ = [
+ '../_base_/datasets/rtts_coco.py', '../_base_/schedules/schedule_1x.py',
+ '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='PAA',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output',
+ num_outs=5),
+ bbox_head=dict(
+ type='PAAHead',
+ reg_decoded_bbox=True,
+ score_voting=True,
+ topk=9,
+ num_classes=5,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ ratios=[1.0],
+ octave_base_scale=8,
+ scales_per_octave=1,
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='GIoULoss', loss_weight=1.3),
+ loss_centerness=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=0.5)),
+ # training and testing settings
+ train_cfg=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.1,
+ neg_iou_thr=0.1,
+ min_pos_iou=0,
+ ignore_iof_thr=-1),
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.6),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# learning rate
+param_scheduler = [
+ dict(
+ type='LinearLR', start_factor=0.001, by_epoch=False, begin=0, end=500),
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=12,
+ by_epoch=True,
+ milestones=[8, 11],
+ gamma=0.1)
+]
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='rtts_detection',
+# name='paa_r50_fpn_1x_rtts',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/rtts_dataset/retinanet_r50_fpn_1x_rtts-coco.py b/configs/detection/rtts_dataset/retinanet_r50_fpn_1x_rtts-coco.py
new file mode 100644
index 0000000..5d6c8ba
--- /dev/null
+++ b/configs/detection/rtts_dataset/retinanet_r50_fpn_1x_rtts-coco.py
@@ -0,0 +1,105 @@
+_base_ = [
+ '../_base_/datasets/rtts_coco.py', '../_base_/schedules/schedule_1x.py',
+ '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='RetinaNet',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_input',
+ num_outs=5),
+ bbox_head=dict(
+ type='RetinaHead',
+ num_classes=5,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ octave_base_scale=4,
+ scales_per_octave=3,
+ ratios=[0.5, 1.0, 2.0],
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0]),
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
+ # model training and testing settings
+ train_cfg=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.4,
+ min_pos_iou=0,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='PseudoSampler'), # Focal loss should use PseudoSampler
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.5),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001),
+ clip_grad=dict(max_norm=35, norm_type=2)) # loss may NaN without clip_grad
+
+param_scheduler = [
+ dict(
+ type='LinearLR', start_factor=0.001, by_epoch=False, begin=0,
+ end=1000),
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=12,
+ by_epoch=True,
+ milestones=[8, 11],
+ gamma=0.1)
+]
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='rtts_detection',
+# name='retinanet_r50_fpn_1x_rtts',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/rtts_dataset/tood_r50_fpn_1x_rtts-coco.py b/configs/detection/rtts_dataset/tood_r50_fpn_1x_rtts-coco.py
new file mode 100644
index 0000000..7f41d2a
--- /dev/null
+++ b/configs/detection/rtts_dataset/tood_r50_fpn_1x_rtts-coco.py
@@ -0,0 +1,95 @@
+_base_ = [
+ '../_base_/datasets/rtts_coco.py', '../_base_/schedules/schedule_1x.py',
+ '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='TOOD',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output',
+ num_outs=5),
+ bbox_head=dict(
+ type='TOODHead',
+ num_classes=5,
+ in_channels=256,
+ stacked_convs=6,
+ feat_channels=256,
+ anchor_type='anchor_free',
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ ratios=[1.0],
+ octave_base_scale=8,
+ scales_per_octave=1,
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ initial_loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ activated=True, # use probability instead of logit as input
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_cls=dict(
+ type='QualityFocalLoss',
+ use_sigmoid=True,
+ activated=True, # use probability instead of logit as input
+ beta=2.0,
+ loss_weight=1.0),
+ loss_bbox=dict(type='GIoULoss', loss_weight=2.0)),
+ train_cfg=dict(
+ initial_epoch=4,
+ initial_assigner=dict(type='ATSSAssigner', topk=9),
+ assigner=dict(type='TaskAlignedAssigner', topk=13),
+ alpha=1,
+ beta=6,
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.6),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='rtts_detection',
+# name='tood_r50_fpn_1x_rtts',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/atss_r50_fpn_1x_ruod.py b/configs/detection/ruod_dataset/atss_r50_fpn_1x_ruod.py
index ac7fbf0..eaa603d 100644
--- a/configs/detection/ruod_dataset/atss_r50_fpn_1x_ruod.py
+++ b/configs/detection/ruod_dataset/atss_r50_fpn_1x_ruod.py
@@ -70,3 +70,17 @@
# optimizer
optim_wrapper = dict(
optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='atss_r50_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/cascade-rcnn_r50_fpn_1x_ruod.py b/configs/detection/ruod_dataset/cascade-rcnn_r50_fpn_1x_ruod.py
index b124cec..4c4bbd9 100644
--- a/configs/detection/ruod_dataset/cascade-rcnn_r50_fpn_1x_ruod.py
+++ b/configs/detection/ruod_dataset/cascade-rcnn_r50_fpn_1x_ruod.py
@@ -188,3 +188,17 @@
score_thr=0.05,
nms=dict(type='nms', iou_threshold=0.5),
max_per_img=100)))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='cascade-rcnn_r50_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/faster-rcnn_r101_fpn_1x_ruod.py b/configs/detection/ruod_dataset/faster-rcnn_r101_fpn_1x_ruod.py
index 4b766c4..825c5c3 100644
--- a/configs/detection/ruod_dataset/faster-rcnn_r101_fpn_1x_ruod.py
+++ b/configs/detection/ruod_dataset/faster-rcnn_r101_fpn_1x_ruod.py
@@ -5,3 +5,17 @@
depth=101,
init_cfg=dict(type='Pretrained',
checkpoint='torchvision://resnet101')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='faster-rcnn_r101_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/faster-rcnn_r50_fpn_1x_ruod.py b/configs/detection/ruod_dataset/faster-rcnn_r50_fpn_1x_ruod.py
index 9a89a75..72c8951 100644
--- a/configs/detection/ruod_dataset/faster-rcnn_r50_fpn_1x_ruod.py
+++ b/configs/detection/ruod_dataset/faster-rcnn_r50_fpn_1x_ruod.py
@@ -117,3 +117,17 @@
# soft-nms is also supported for rcnn testing
# e.g., nms=dict(type='soft_nms', iou_threshold=0.5, min_score=0.05)
))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='faster-rcnn_r50_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/faster-rcnn_r50_fpn_2x_ruod.py b/configs/detection/ruod_dataset/faster-rcnn_r50_fpn_2x_ruod.py
index 9a89a75..21801b4 100644
--- a/configs/detection/ruod_dataset/faster-rcnn_r50_fpn_2x_ruod.py
+++ b/configs/detection/ruod_dataset/faster-rcnn_r50_fpn_2x_ruod.py
@@ -117,3 +117,17 @@
# soft-nms is also supported for rcnn testing
# e.g., nms=dict(type='soft_nms', iou_threshold=0.5, min_score=0.05)
))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='faster-rcnn_r50_fpn_2x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/faster-rcnn_x101-32x4d_fpn_1x_ruod.py b/configs/detection/ruod_dataset/faster-rcnn_x101-32x4d_fpn_1x_ruod.py
index d3cd0fe..2d3e31a 100644
--- a/configs/detection/ruod_dataset/faster-rcnn_x101-32x4d_fpn_1x_ruod.py
+++ b/configs/detection/ruod_dataset/faster-rcnn_x101-32x4d_fpn_1x_ruod.py
@@ -13,3 +13,17 @@
style='pytorch',
init_cfg=dict(
type='Pretrained', checkpoint='open-mmlab://resnext101_32x4d')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='faster-rcnn_x101-32x4d_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/faster-rcnn_x101-64x4d_fpn_1x_ruod.py b/configs/detection/ruod_dataset/faster-rcnn_x101-64x4d_fpn_1x_ruod.py
index 0e12094..f60c5af 100644
--- a/configs/detection/ruod_dataset/faster-rcnn_x101-64x4d_fpn_1x_ruod.py
+++ b/configs/detection/ruod_dataset/faster-rcnn_x101-64x4d_fpn_1x_ruod.py
@@ -13,3 +13,17 @@
style='pytorch',
init_cfg=dict(
type='Pretrained', checkpoint='open-mmlab://resnext101_64x4d')))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='faster-rcnn_x101-64x4d_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/fcos_r50-caffe_fpn_gn-head_1x_ruod.py b/configs/detection/ruod_dataset/fcos_r50-caffe_fpn_gn-head_1x_ruod.py
index 259f4ec..5205176 100644
--- a/configs/detection/ruod_dataset/fcos_r50-caffe_fpn_gn-head_1x_ruod.py
+++ b/configs/detection/ruod_dataset/fcos_r50-caffe_fpn_gn-head_1x_ruod.py
@@ -75,3 +75,17 @@
optimizer=dict(lr=0.01),
paramwise_cfg=dict(bias_lr_mult=2., bias_decay_mult=0.),
clip_grad=dict(max_norm=35, norm_type=2)) # loss may NaN without clip_grad
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='fcos_r50-caffe_fpn_gn_head_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/paa_r50_fpn_1x_ruod.py b/configs/detection/ruod_dataset/paa_r50_fpn_1x_ruod.py
new file mode 100644
index 0000000..002b790
--- /dev/null
+++ b/configs/detection/ruod_dataset/paa_r50_fpn_1x_ruod.py
@@ -0,0 +1,107 @@
+_base_ = [
+ '../_base_/datasets/ruod_coco_detection.py',
+ '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(
+ type='PAA',
+ data_preprocessor=dict(
+ type='DetDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True,
+ pad_size_divisor=32),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ norm_cfg=dict(type='BN', requires_grad=True),
+ norm_eval=True,
+ style='pytorch',
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ start_level=1,
+ add_extra_convs='on_output',
+ num_outs=5),
+ bbox_head=dict(
+ type='PAAHead',
+ reg_decoded_bbox=True,
+ score_voting=True,
+ topk=9,
+ num_classes=10,
+ in_channels=256,
+ stacked_convs=4,
+ feat_channels=256,
+ anchor_generator=dict(
+ type='AnchorGenerator',
+ ratios=[1.0],
+ octave_base_scale=8,
+ scales_per_octave=1,
+ strides=[8, 16, 32, 64, 128]),
+ bbox_coder=dict(
+ type='DeltaXYWHBBoxCoder',
+ target_means=[.0, .0, .0, .0],
+ target_stds=[0.1, 0.1, 0.2, 0.2]),
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='GIoULoss', loss_weight=1.3),
+ loss_centerness=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=0.5)),
+ # training and testing settings
+ train_cfg=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.1,
+ neg_iou_thr=0.1,
+ min_pos_iou=0,
+ ignore_iof_thr=-1),
+ allowed_border=-1,
+ pos_weight=-1,
+ debug=False),
+ test_cfg=dict(
+ nms_pre=1000,
+ min_bbox_size=0,
+ score_thr=0.05,
+ nms=dict(type='nms', iou_threshold=0.6),
+ max_per_img=100))
+
+# optimizer
+optim_wrapper = dict(
+ optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# learning rate
+param_scheduler = [
+ dict(
+ type='LinearLR', start_factor=0.001, by_epoch=False, begin=0, end=500),
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=12,
+ by_epoch=True,
+ milestones=[8, 11],
+ gamma=0.1)
+]
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='paa_r50_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/retinanet_r50_fpn_1x_ruod.py b/configs/detection/ruod_dataset/retinanet_r50_fpn_1x_ruod.py
index faba249..89405fc 100644
--- a/configs/detection/ruod_dataset/retinanet_r50_fpn_1x_ruod.py
+++ b/configs/detection/ruod_dataset/retinanet_r50_fpn_1x_ruod.py
@@ -89,3 +89,17 @@
milestones=[8, 11],
gamma=0.1)
]
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='retinanet_r50_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/ssd300_120e_ruod.py b/configs/detection/ruod_dataset/ssd300_120e_ruod.py
index f189bf8..75cfa32 100644
--- a/configs/detection/ruod_dataset/ssd300_120e_ruod.py
+++ b/configs/detection/ruod_dataset/ssd300_120e_ruod.py
@@ -85,3 +85,17 @@
# USER SHOULD NOT CHANGE ITS VALUES.
# base_batch_size = (8 GPUs) x (8 samples per GPU)
auto_scale_lr = dict(base_batch_size=64)
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='ssd300_120e_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/ssd512_120e_ruod.py b/configs/detection/ruod_dataset/ssd512_120e_ruod.py
index ae6db55..6a9590a 100644
--- a/configs/detection/ruod_dataset/ssd512_120e_ruod.py
+++ b/configs/detection/ruod_dataset/ssd512_120e_ruod.py
@@ -58,3 +58,17 @@
# USER SHOULD NOT CHANGE ITS VALUES.
# base_batch_size = (8 GPUs) x (8 samples per GPU)
auto_scale_lr = dict(base_batch_size=64)
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='ssd512_120e_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/ruod_dataset/tood_r50_fpn_1x_ruod.py b/configs/detection/ruod_dataset/tood_r50_fpn_1x_ruod.py
index 0ed81d5..57b44f0 100644
--- a/configs/detection/ruod_dataset/tood_r50_fpn_1x_ruod.py
+++ b/configs/detection/ruod_dataset/tood_r50_fpn_1x_ruod.py
@@ -79,3 +79,17 @@
# optimizer
optim_wrapper = dict(
optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))
+
+# add WandbVisBackend
+# vis_backends = [
+# dict(type='LocalVisBackend'),
+# dict(type='WandbVisBackend',
+# init_kwargs=dict(
+# project='RUOD_detection',
+# name='tood_r50_fpn_1x_ruod',
+# entity='lqit',
+# )
+# )
+# ]
+# visualizer = dict(
+# type='DetLocalVisualizer', vis_backends=vis_backends, name='visualizer')
diff --git a/configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_urpc-coco.py b/configs/detection/uod_air/faster-rcnn_r50_fpn_1x_urpc-coco.py
similarity index 98%
rename from configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_urpc-coco.py
rename to configs/detection/uod_air/faster-rcnn_r50_fpn_1x_urpc-coco.py
index c9baaa3..2cb93f6 100644
--- a/configs/detection/duo_dataset/faster-rcnn_r50_fpn_1x_urpc-coco.py
+++ b/configs/detection/uod_air/faster-rcnn_r50_fpn_1x_urpc-coco.py
@@ -1,5 +1,5 @@
_base_ = [
- '../_base_/datasets/urpc-2020_coco_detection.py',
+ '../_base_/datasets/urpc2020/urpc2020-validation_coco_detection.py',
'../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
]
diff --git a/configs/detection/duo_dataset/retinanet_r50_fpn_1x_urpc-coco.py b/configs/detection/uod_air/retinanet_r50_fpn_1x_urpc-coco.py
similarity index 97%
rename from configs/detection/duo_dataset/retinanet_r50_fpn_1x_urpc-coco.py
rename to configs/detection/uod_air/retinanet_r50_fpn_1x_urpc-coco.py
index 1113e34..eef8d7a 100644
--- a/configs/detection/duo_dataset/retinanet_r50_fpn_1x_urpc-coco.py
+++ b/configs/detection/uod_air/retinanet_r50_fpn_1x_urpc-coco.py
@@ -1,5 +1,5 @@
_base_ = [
- '../_base_/datasets/urpc-2020_coco_detection.py',
+ '../_base_/datasets/urpc2020/urpc2020-validation_coco_detection.py',
'../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
]
diff --git a/configs/detection/urpc2020_dataset/README.md b/configs/detection/urpc2020_dataset/README.md
new file mode 100644
index 0000000..e074880
--- /dev/null
+++ b/configs/detection/urpc2020_dataset/README.md
@@ -0,0 +1,69 @@
+# Detecting Underwater Objects
+
+> [Detecting Underwater Objects](https://arxiv.org/abs/2106.05681)
+> [Underwater Robot Professional Contest 2020](https://www.heywhale.com/home/competition/5e535a612537a0002ca864ac/content/0)
+
+
+
+Underwater object detection for robot picking has attracted a lot of interest. However, it is still an unsolved problem due to several challenges. We take steps towards making it more realistic by addressing the following challenges. Firstly, the currently available datasets basically lack the test set annotations, causing researchers must compare their method with other SOTAs on a self-divided test set (from the training set). Training other methods lead to an increase in workload and different researchers divide different datasets, resulting there is no unified benchmark to compare the performance of different algorithms. Secondly, these datasets also have other shortcomings, e.g., too many similar images or incomplete labels. Towards these challenges we introduce a dataset, Detecting Underwater Objects (DUO), and a corresponding benchmark, based on the collection and re-annotation of all relevant datasets. DUO contains a collection of diverse underwater images with more rational annotations. The corresponding benchmark provides indicators of both efficiency and accuracy of SOTAs (under the MMDtection framework) for academic research and industrial applications, where JETSON AGX XAVIER is used to assess detector speed to simulate the robot-embedded environment.
+
+
+
+
+
+