This directory contains the model code for Temporal Alignment Networks and the model code for backbone S3D-word2vec.
Preparation for end-to-end training
If you want to run experiments on S3D, you need to download the pretrained weight for S3D from MIL-NCE paper.
cd s3d_milnce/
wget https://www.rocq.inria.fr/cluster-willow/amiech/howto100m/s3d_howto100m.pth
wget https://www.rocq.inria.fr/cluster-willow/amiech/howto100m/s3d_dict.npy
So your directory should be like:
model/
s3d_milnce/
__init__.py
s3d_dict.npy
s3d_howto100m.pth
s3dg.py
...