the number of frames in a clip is set to be 8 in data_preparation_ActivityNet/generate_nolabel.py

[winnechan]

Hi, I found that the number of frames in a clip is set to be 8 in data_preparation_ActivityNet/generate_nolabel.py while in C3D feature it should be 16, does this would affect the normalized location labels and the fetched features from sub_activitynet_v1-3.c3d.hdf5 for training ?

[yytzsy]

In sub_activitynet_v1-3.c3d.hdf5, they extract video clip features with 16 frames in a clip, but the stride is 8. That is to say, [0,16], [8,24], [16,32] ... are extracted