The video-cnn-feat toolbox provides python code and scripts for extracting CNN features from video frames by pre-trained MXNet models. We have used this toolbox for our winning solution at TRECVID 2018 ad-hoc video search (AVS) task and in our W2VV++ paper.
