ResNet-Video-Features

Temporal video features extracted from ImageNet pre-trained ResNet-152.

---

Extract Frames

Firstly, we need to extract the frames of each video by FFmpeg.

• On Linux, please refer to the extract_frames.sh file.
• On Windows, the scripts are similar:

# Make a folder for each video
for %f in (./Videos/*.mp4) do mkdir "./Frames_folders/%~nf"

# Extract and save the frames of each video
for %f in (./Videos/*.mp4) do ffmpeg -i "./Videos/%f" "./Frames_folders/%~nf/%05d.jpg"

Extract Featrues for Frames

Just run the resnet_features.py in Python:

python resnet_features.py

Requisites:

Pytorch >= 0.4.1