3 ms·
VQVAE is trained to reconstruct the image, so in theory it should contain all the information (both content and location) inside its embeddings.
by rafaelero 2y ago
VQVAE is trained to reconstruct the image, so in theory it should contain all the information (both content and location) inside its embeddings.