Jian
Jian
@YaaYang You must follow https://github.com/yjxiong/temporal-segment-networks for video classification. Then with these classification models, you can extract features by using the FC layer. 
@YaaYang rescale 是额外做的。一个视频抽完特征大小NX400，再rescale到100X400。
 This layer ,200X2=400, 200 flow feature, 200rgb feature , 200 is also activitynet class number.
Hi, you can use our inference code and weights for evaluating our paper results. train.py will be released later. Keep your attention.
train.py will be released later.
Hi，we are planning to release a much more real-time model.But the the detection performance may be lower
We just use WIDERFACE training dataset from http://shuoyang1213.me/WIDERFACE/
@nyck33 I am sure we only use WIDERFACE datasset and didnot append other dataset.
gpu
Hi @linzhi123 We just released inference code. Training code is temporarily unavailable.