bornfly-detachment / asymmetric_magvitv2

In 2024, the strongest open-source implementation of asymmetric magvit_v2 supports inference code but excludes VQVAE. It supports the joint encoding of images and videos, accommodating arbitrary video lengths and resolutions. It surpasses all open-source models in FID and FVD, with 4z and 16z models available on huggingface.
134Updated 5 months ago

Alternatives and similar repositories for asymmetric_magvitv2:

Users that are interested in asymmetric_magvitv2 are comparing it to the libraries listed below