search

mysee1989 / TCAE

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

The dataset #9

Closed yy19931029 closed 4 years ago

yy19931029 commented 4 years ago

hello, I download the dataset voxceleb to train the model, that is about 1600 people, but the result of val error was about 0.5, maybe it was to high. It's my dataset not enough or other reason?

mysee1989 commented 4 years ago

You should download the voxceleb1 and voxceleb2 datasets, which contain about 7000 subjects.