Sign Language Poses Dataset

[syntheticfinds]

Would it be possible to share the dataset of sign language poses that's used during pre-training and fine-tuning? I understand that the raw version is more than 500 GB, so perhaps, the quantized version can be shared? Thank you in advance for your help!

[AmitMY]

Unfortunately not possible to share the source 500GB.
I could share the quantized version done with (https://github.com/sign-language-processing/sign-vq) but from my experiments, the quantization is not good enough, and therefore the model does not learn well. The translation model available in this repo is not good.

I think a more promising approach at the moment would be diffusion. Using only a few thousand videos, without pre-training, we achieved great results https://github.com/rotem-shalev/Ham2Pose, and for diffusion one could use pretrained models