revert

custum_3d_diffusion/custum_pipeline/unifield_pipeline_img2mvimg.py

@@ -204,8 +204,7 @@ class StableDiffusionImage2MVCustomPipeline(
         #     batch_size = len(image)
         # else:
         #     batch_size = image.shape[0]
-        # device = self._execution_device
-        device = "cuda"
+        device = self._execution_device
         # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
         # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
         # corresponds to doing no classifier free guidance.
@@ -214,9 +213,7 @@ class StableDiffusionImage2MVCustomPipeline(
         # 3. Encode input image
         emb_image = image
         
-        image_embeddings = self._encode_image(emb_image, device, num_images_per_prompt, do_classifier_free_guidance).to(device=self.unet.device, dtype=self.unet.dtype)
-        print("DEBUG: image_embeddings", image_embeddings.dtype, image_embeddings.device)
-        print("DEBUG: version v111")
+        image_embeddings = self._encode_image(emb_image, device, num_images_per_prompt, do_classifier_free_guidance)
         cond_latents = self.encode_latents(image, image_embeddings.device, image_embeddings.dtype, height_cond, width_cond)
         cond_latents = torch.cat([torch.zeros_like(cond_latents), cond_latents]) if do_classifier_free_guidance else cond_latents
         image_pixels = self.feature_extractor(images=emb_image, return_tensors="pt").pixel_values