code of CMU entrance at Surveillance Event Detection TRECVID2017
depends on my personal tensorflow experiment framework and python toolkit
- TrackDb: file system based database storing person tracking data
- ClipDb: file system based database storing video clip data
- FtDb: file system based database storing feature data
- crop_instant_ft_in_track: crop feature at each time unit from the person trajectory
- crop_duration_ft_in_track: crop the feature from the person trajectory within the given time interval
- crop_clip_in_track: crop the trajectory from the video clip
- netvlad: NetVLAD pooling
- attention: self-attention on temporal sequence
- attention.py: self-attention on temporal sequence
- netvlad.py: NetVLAD pooling
- svm: vanilla SVM classifier
- c3d_lj.py: C3D feature extraction
- paf_merge.py: Part Affinity Field of person pose feature
- twostream.py: RGB stream and flow stream feature
- vlad.py: VLAD pooling of features