search

mysee1989 / TCAE

Self-supervised Representation Learning from Videos for Facial Action Unit Detection
162 stars 27 forks source link

Question about directory structure #10

Open sidarthd opened 4 years ago

sidarthd commented 4 years ago

'data/id09238/VqEJCd7pbgQ/0_folder/img_0_001.jpg'

What is '0_folder' here? Is it the frame number? If so how are you sampling faces of the same person from the video?

mysee1989 commented 4 years ago

0_folder is the directory that I created to store facial images extracted from a single video clip.

Since each folder only contains frames/images extracted from a single video clip, then it is convenient to sample arbitrary facial image pairs.

sidarthd commented 4 years ago

How did you get the timestamps and durations of clips in a video?

mysee1989 commented 4 years ago

During training, there is no need to get the timestamps or durations of clips in a video.

Before training, I extracted and save the facial frames at 1 fps from each video. Each face was detected & aligned.

sidarthd commented 4 years ago

Okay, did you save all of the faces in one video in a single folder ?

mysee1989 commented 4 years ago

Yes. The facial frames from a video should be saved in a single folder.

enginbaglayici commented 3 years ago

Hi,

I am also trying to understand directory structure but I couldn't figure out why there are 4 folders ('data/id09238/VqEJCd7pbgQ/0_folder/img_0_001.jpg') As far as I understand: id09238 means a subject. VqEJCd7pbgQ - that I couldnt understand 0_folder - belongs to a video of the subject

Thanks for the help :)