how can i get 1024 dimensions of each frame? about video_feature_extractor HOT 8 CLOSED

It is the global average pooled output that you have just before the last linear layer.

from video_feature_extractor.

Is ffprobe working on these videos ? Maybe to be sure that the path is correct, please replace the path with absolute path instead of relative path ?

from video_feature_extractor.

Why would you need 1024 ? The feature dimension for ResNet and ResNext is 2048.

from video_feature_extractor.

What exactly do these 2048 dimensions represent?

from video_feature_extractor.

so, in a sense, if you have 1024 dimensions you're interested in, @linjingxu, you could just add a layer of 1024 dimensions on the 2048 outputs, assuming you have labeled your video data in 1024 dims..

from video_feature_extractor.

Could you help me to look at this problem? This is my debugger's result. Why in "loader" i didn't get any video tensor???? decoding video failed

from video_feature_extractor.

This is my .csv file. What should i do with this ?? Is there any problems?

from video_feature_extractor.

Is ffprobe working on these videos ? Maybe to be sure that the path is correct, please replace the path with absolute path instead of relative path ?

thanks i got it！！

from video_feature_extractor.