ERRATA (ECCV paper, Figure 2): first convolutional block of each ResNet unit is 1x1, not 3x3. The code is correct.
If you use this code as part of any published research, please acknowledge the following paper:
@InProceedings{Garcia_2018_ECCV,
author = {Garcia, Nuno C. and Morerio, Pietro and Murino, Vittorio},
title = {Modality Distillation with Multiple Stream Networks for Action Recognition},
booktitle = {The European Conference on Computer Vision (ECCV)},
month = {September},
year = {2018}
}
cat ./log/uwa3dii/s1_train_depth_01012018_010101__dset_uwa3dii_eval_mode_cross_view/log.txt
depth:
python s1_train_stream.py --dset=ntu --modality=depth --eval=cross_subj
RGB:
python s1_train_stream.py --dset=nwucla --modality=rgb
Define the right path for depth and rgb checkpoints in s2_twostream_depth_rgb.py
python s2_twostream_depth_rgb.py --interaction --dset=ntu-mini
python s2_twostream_depth_rgb --dset=uwa3dii --just_eval --ckpt=./step2_checkpoint_dir_model.ckpt
python s3_distillation.py --dset=nwucla --ckpt=./step2_checkpoint_dir_model.ckpt
Define the right path for depth (checkpoint from step 3) and rgb (checkpoint from step 2) ckpts in s4_depth_hall.py
python s4_depth_hall.py --dset=uwa3dii
python s4_depth_hall.py --dset=uwa3dii --just_eval --ckpt=./step4_checkpoint_dir_model.ckpt