conda install pytorch torchvision cuda80 -c soumith scikit-image h5py
- FFmpeg, FFprobe
wget http://johnvansickle.com/ffmpeg/releases/ffmpeg-release-64bit-static.tar.xz
tar xvf ffmpeg-release-64bit-static.tar.xz
cd ./ffmpeg-3.3.3-64bit-static/; sudo cp ffmpeg ffprobe /usr/local/bin;
- Python 3
- Download this code.
- Download pretrained model 'resnet152.pth' and 'resnext-101-64f-kinetics.pth'. Then put them in the main folder.
- Download the images, unzip and put the 'visual_data' folder in the main folder.
Run
python main_img_train.py --mode feature
to get motion features of train set
Run
python main_img_val.py --mode feature
to get motion features of validation set
Run
python prepro_feats_train.py
to get visual features of train set Run
python prepro_feats_val.py
to get visual features of validation set