MCPcopy Create free account
hub / github.com/MiniMax-AI/VTP / decode_to_images

Method decode_to_images

generation/tokenizer/vtp_tokenizer.py:95–111  ·  view source on GitHub ↗
(self, z)

Source from the content-addressed store, hash-verified

93 return latents.detach().cpu()
94
95 def decode_to_images(self, z):
96 with torch.no_grad():
97 if not z.is_cuda:
98 z = z.cuda()
99
100 B, C, H_latent, W_latent = z.shape
101 self._current_img_h = H_latent * self.patch_size
102 self._current_img_w = W_latent * self.patch_size
103
104 decoded = self.model.get_latents_decoded_images(z)
105
106 # Apply inverse normalization to get [0, 1] range
107 decoded = self.transform_inv(decoded)
108 # Convert to [0, 255] uint8
109 images = torch.clamp(decoded * 255, 0, 255)
110 images = images.permute(0, 2, 3, 1).to("cpu", dtype=torch.uint8).numpy()
111 return images

Callers

nothing calls this directly

Calls 1

Tested by

no test coverage detected