push

inference/cli_demo_camera_i2v_pcd.py

@@ -498,7 +498,6 @@ if __name__ == "__main__":
     #  "--upscale", str(upscale),
     #     "--upscale_factor", str(upscale_factor),
     #     "--refine", str(refine),
-    
 
     args = parser.parse_args()
     dtype = torch.float16 if args.dtype == "float16" else torch.bfloat16

inference/utils.py

@@ -168,7 +168,8 @@ def load_sd_upscale(ckpt, inf_device):
     return out
 
 
-def upscale(upscale_model, tensor: torch.Tensor, inf_device, output_device="cpu") -> torch.Tensor:
+def upscale(upscale_model, tensor: torch.Tensor, inf_device, output_device="cpu", upscale_factor) -> torch.Tensor:
+
     memory_required = module_size(upscale_model.model)
     memory_required += (
         (512 * 512 * 3) * tensor.element_size() * max(upscale_model.scale, 1.0) * 384.0
@@ -185,14 +186,14 @@ def upscale(upscale_model, tensor: torch.Tensor, inf_device, output_device="cpu"
     )
 
     pbar = ProgressBar(steps, desc="Tiling and Upscaling")
-
+    # upscale_model.scale
     s = tiled_scale(
         samples=tensor.to(torch.float16),
         function=lambda a: upscale_model(a),
         tile_x=tile,
         tile_y=tile,
         overlap=overlap,
-        upscale_amount=upscale_model.scale,
+        upscale_amount=upscale_factor,
         pbar=pbar,
     )
 
@@ -204,7 +205,7 @@ def upscale_batch_and_concatenate(upscale_model, latents, inf_device, output_dev
     upscaled_latents = []
     for i in range(latents.size(0)):
         latent = latents[i]
-        upscaled_latent = upscale(upscale_model, latent, inf_device, output_device)
+        upscaled_latent = upscale(upscale_model, latent, inf_device, output_device, upscale_factor)
         upscaled_latents.append(upscaled_latent)
     return torch.stack(upscaled_latents)