Environment
python 3.9.7
torch 2.0.0
torchvision 0.15.1
numpy 1.20.3
To train the model:
python train.py -root /data/E2H_source/ -mode rgb -save_model I3D_RGB_HMDB -dset H -CLASS_NUM 13 -trainfile ./targetlistname_My_H.txt -r1 0.2 -r2 0.3 -th 0.9 -lr 0.025 -sset E -droot /data/HMDB51-frame -PRE 5 -method Pro -r3 0.5
where /data/E2H_source/ is the directory containing synthesized video frames and /data/HMDB51-frame is directory containing extracted video frames from HMDB dataset. The synthesized video frames of E2H can be found in https://drive.google.com/file/d/1D2vuK7a1gXig_hvwOuvo0LUEEbkCEzlc/view?usp=sharing (placing at start) and https://drive.google.com/file/d/1m2_C2C-7b-geoD46W9Mh2Kdtd_A6X15Y/view?usp=sharing (placing at middle) .
The rgb_imagenet.pt can be found in https://github.com/piergiaj/pytorch-i3d/tree/master/models.