README.md

TSN-Pytorch

Now in experimental release, suggestions welcome.

Note: always use git clone --recursive https://github.com/yjxiong/tsn-pytorch to clone this project. Otherwise you will not be able to use the inception series CNN archs.

This is a reimplementation of temporal segment networks (TSN) in PyTorch. All settings are kept identical to the original caffe implementation.

For optical flow extraction and video list generation, you still need to use the original TSN codebase.

Training

To train a new model, use the main.py script.

The command to reproduce the original TSN experiments of RGB modality on UCF101 can be

python main.py ucf101 RGB <ucf101_rgb_train_list> <ucf101_rgb_val_list> \
   --arch BNInception --num_segments 3 \
   --gb 20 --lr 0.001 --lr_steps 30 60 --epochs 80 \
   -b 128 -j 8 \
   --snapshot_pref ucf101_bninception_ 

For flow models:

python main.py ucf101 Flow <ucf101_flow_train_list> <ucf101_flow_val_list> \
   --arch BNInception --num_segments 3 \
   --gb 20 --lr 0.001 --lr_steps 190 300 --epochs 340 \
   -b 128 -j 8 \
   --snapshot_pref ucf101_bninception_ --flow_pref flow_  

For RGB-diff models:

python main.py ucf101 RGBDiff <ucf101_rgb_train_list> <ucf101_rgb_val_list> \
   --arch BNInception --num_segments 7 \
   --gb 40 --lr 0.001 --lr_steps 80 160 --epochs 180 \
   -b 128 -j 8 \
   --snapshot_pref ucf101_bninception_ 

Testing

After training, there will checkpoints saved by pytorch, for example ucf101_bninception_rgb_checkpoint.pth.

Use the following command to test its performance in the standard TSN testing protocol:

python test_models.py ucf101 RGB <ucf101_rgb_val_list> ucf101_bninception_rgb_checkpoint.pth \
   --arch BNInception --save_scores <score_file_name>

Or for flow models:

python test_models.py ucf101 Flow <ucf101_rgb_val_list> ucf101_bninception_flow_checkpoint.pth \
   --arch BNInception --save_scores <score_file_name> --flow_pref flow_