search

hx173149 / C3D-tensorflow

C3D is a modified version of BVLC tensorflow to support 3D ConvNets.
MIT License
588 stars 265 forks source link

"a 8-frame overlap" #97

Open antoniomagnani opened 5 years ago

antoniomagnani commented 5 years ago

3.We follow the same way to extract clips from video as the C3D paper saying:'To extract C3D feature, a video is split into 16 frame long clips with a 8-frame overlap between two consecutive clips.These clips are passed to the C3D network to extract fc6 activations. These clip fc6 activations are averaged to form a 4096-dim video descriptor which is then followed by an L2-normalization' <

I think is not true cause you simply take a random index for the first frame from a certain sequence. After the firsts 16 frames are taken and a new sequence is considered. There is no mention of 50% overlapping in this code.

Naegionn commented 5 years ago

The overlapp is used for classification on video level, while the provided predict function is clip level