search

tub-cv-group / conclugen

Official repository for our CVPR 2024 Workshop paper "Multi-Task Multi-Modal Self-Supervised Learning for Facial Expression Recognition".
16 stars 2 forks source link

Extracting image features about facial expressions #2

Open laodar opened 5 months ago

laodar commented 5 months ago

Thanks for your excellent work. I want to extract features about expressions from images, which pretrained model is prefered?

florianblume commented 5 months ago

Hi, thank you for your interest in our work! Could you elaborate a bit further what exactly you want to do? You mean by pretrained model which one of ours?

laodar commented 5 months ago

Hi, thank you for your interest in our work! Could you elaborate a bit further what exactly you want to do? You mean by pretrained model which one of ours?

I want to extract features only from image (or video) signals and use them to calculate a 'perceptual loss' that measures how close a generated image is to the ground truth image in some 'expression space'. I noticed that you trained models with many variants. Now I think model_name=contrast_only_video_audio will be prefered in my task, then how can I obtain the pretrained image feature extractor and build a 'perceptual loss' ?