enhancing-transformers
enhancing-transformers copied to clipboard
Incomplete implementation of RQ-VAE
It seems that the ViTVQ only consider the origion VQVAE(Not considering codes has 3 dimensions, e.g. HxWxN, N means residual times in RQVAE) https://github.com/thuanz123/enhancing-transformers/blob/1778fc497ea11ed2cef134404f99d4d6b921cda9/enhancing/modules/stage1/vitvqgan.py#L81
RQ-VAE is only mentioned here(and flag use_residual
is always False)
https://github.com/thuanz123/enhancing-transformers/blob/1778fc497ea11ed2cef134404f99d4d6b921cda9/enhancing/modules/stage1/quantizers.py#L42
Did I miss something? Looking forward to your reply!