feature_kit

(Video) Feature Extraction Code for Visual Abductive Reasoning

Partially forked code from anet2016-cuhk-feature and caffemodel2pytorch. See also here (CUHK & ETH & SIAT Solution to ActivityNet Challenge 2016) for more details.

Feature extraction

• RGB ResNet-200 feature
• Optical flow BN-Inception feature

Brief instruction

1. Download pretrained models:

bash models/get_reference_models.sh

2. Extract feature through:

python extract_feature.py 

Note: See more configurations in extract_feature.py.