porntech

/

sex-position

like 49

If not, could you talk about what you did for dataset curation (resolution, image count per class, hand-labelled or BLIP, etc)?

Hi, thanks for you interest. You can follow the instruction of hf to run the inference, for the dataset, I crawled large amount of image(more than 100k for each class), I did't annotate them with my self instead use the tags for the image as label so the labels are noisy and the performance is not that good. I'm currently focusing on action classification, if action classification can achieve good accuracy then I will further improve the image classification model