training_vlog.sh

#!/usr/bin/env bash

# Envs
# source activate pytorch-0.4.0

# Pythonpath
PYTHONPATH=.

# Settings
resume=/home/fbaradel/log_eccv18
root=$VLOG

# Train the object head only with f=MLP
epochs=10
heads=object
python main.py --resume $resume \
--root $root \
--blocks 2D_2D_2D_2.5D \
--object-head 2D \
--add-background \
--train-set train+val \
--arch orn_two_heads \
--depth 50 \
-t 4 \
-b 32 \
--cuda \
--dataset vlog \
--heads $heads \
--epochs $epochs \
--pooling avg \
--print-freq 100 \
--pooling avg \
--mask-confidence 0.5 \
-j 4

# Train the two heads with f=RNN and pooling is avg for context head
epochs=10
heads=object+context
python main.py --resume $resume \
--root $root \
--blocks 2D_2D_2D_2.5D \
--object-head 2D \
--add-background \
--train-set train+val \
--arch orn_two_heads \
--depth 50 \
-t 4 \
-b 8 \
--cuda \
--dataset vlog \
--heads $heads \
--epochs $epochs \
--print-freq 100 \
--pooling rnn \
--mask-confidence 0.5 \
-j 4

# Finally validate on the test set
epochs=10
heads=object+context
python main.py --resume $resume \
--root $root \
--blocks 2D_2D_2D_2.5D \
--object-head 2D \
--add-background \
--train-set train+val \
--arch orn_two_heads \
--depth 50 \
-t 4 \
-b 10 \
--cuda \
--dataset vlog \
--heads $heads \
--epochs $epochs \
--print-freq 100 \
-j 4 \
--pooling rnn \
--nb-crops 8 \
--mask-confidence 0.5 \
-e