Home

This is the documentation for video_features. A library (😅) that allows you to extract features from raw videos using the pre-trained nets. So far, it supports several extractors that capture visual appearance, calculates optical flow, and, even, audio features. The source code lives at v-iashin/video_features.

The source code was intended to support the feature extraction pipeline for two of my papers (BMT and MDVC). This library (😅) somehow emerged out of that code and now has more models implemented.

---

Supported models

If you would like to see more, please create an Issue.

Action Recognition

• I3D-Net RGB + Flow (Kinetics 400)
• R(2+1)d RGB (Kinetics 400)

Sound Recognition

• VGGish (AudioSet)

Optical Flow

• RAFT (FlyingChairs, FlyingThings3D, Sintel)
• PWC-Net (Sintel)

Image Recognition

• ResNet-50 (ImageNet)