This page holds a reference for example configs, pretrained models and training/evaluation metrics. You can access these models from code using d2go.model_zoo API.
-
Get pretrained models in python:
from d2go.model_zoo import model_zoo model = model_zoo.get('faster_rcnn_fbnetv3a_C4.yaml', trained=True)
-
Train: the "name" column contains a link to the config file. Running
d2go.train_net --config-file
with the config file will reproduce the corresponding model. -
Evaluation: Running
d2go.train_net --config-file path/to/the/config --eval-only MODEL.WEIGHTS path/to/the/model/weights
with the config file and pretrained model will evaluate the results. See details in Getting Started. -
Training curves and other statistics can be found in
metrics
for each model.
FBNet series are efficient mobile backbones discovered via neural architecture search, which are specially optimized for mobile devices. Please see details in the paper. If using our code/models in your research, please cite our paper:
@article{dai2020fbnetv3,
title={FBNetV3: Joint architecture-recipe search using neural acquisition function},
author={Dai, Xiaoliang and Wan, Alvin and Zhang, Peizhao and Wu, Bichen and He, Zijian and Wei, Zhen and Chen, Kan and Tian, Yuandong and Yu, Matthew and Vajda, Peter and others},
journal={arXiv preprint arXiv:2006.02049},
year={2020}
}
name | box AP | latency* | model id | download |
---|---|---|---|---|
Faster-RCNN-FBNetV3A | 22.99 | 73.3ms | 246823121 | model |metrics |
Faster-RCNN-FBNetV3A-dsmask | 21.06 | 33.7ms | 250414811 | model |metrics |
Faster-RCNN-FBNetV3G-FPN | 43.13 | - | 250356938 | model |metrics |
*: tested on Samsung Galaxy S8 with quantization |
name | box AP | mask AP | model id | download |
---|---|---|---|---|
Mask-RCNN-FBNetV3A | 23.05 | 20.71 | 268421013 | model |metrics |
Mask-RCNN-FBNetV3A-dsmask | 21.76 | 19.97 | 268412271 | model |metrics |
Mask-RCNN-FBNetV3G-FPN | 43.31 | 39.24 | 287445123 | model |metrics |
name | box AP | kp. AP | model id | download |
---|---|---|---|---|
Keypoint-RCNN-FBNetV3A-dsmask | 31.24 | 35.56 | 250430934 | model |metrics |