asdasd

app.py

@@ -12,6 +12,8 @@ import cv2
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
 # Inpainting pipeline
+
+
 unet = UNet2DConditionModel.from_pretrained("pablodawson/ldm3d-inpainting", cache_dir="cache", subfolder="unet", in_channels=9, low_cpu_mem_usage=False, ignore_mismatched_sizes=True)
 pipe = StableDiffusionLDM3DInpaintPipeline.from_pretrained("Intel/ldm3d-4c", cache_dir="cache" ).to(device)
 
@@ -50,7 +52,7 @@ def estimate_depth(image):
 
     output = prediction.cpu().numpy()
 
-    output= 255 * output/np.max(output)
+    output= 255 * (output - np.min(output))/(np.max(output) - np.min(output))
     
     return Image.fromarray(output.astype("uint8"))
 
@@ -70,11 +72,12 @@ def predict(dict, depth, prompt="", negative_prompt="", guidance_scale=7.5, step
 
     init_image = cv2.resize(dict["image"], (512, 512))
     
-    if (depth == None):
+    if (depth is None):
         depth_image = estimate_depth(init_image)
     else:
         depth_image = depth
-    
+        depth_image = Image.fromarray(depth_image[:,:,0].astype("uint8"))
+
     scheduler = getattr(diffusers, scheduler_class_name)
     pipe.scheduler = scheduler.from_pretrained("Intel/ldm3d-4c", subfolder="scheduler")
     
@@ -130,7 +133,7 @@ with image_blocks as demo:
                 with gr.Column():
                     image = gr.Image(source='upload', tool='sketch', elem_id="image_upload", type="numpy", label="Upload",height=400)
                     depth = gr.Image(source='upload', elem_id="depth_upload", type="numpy", label="Upload",height=400)
-
+                    
                     with gr.Row(elem_id="prompt-container", mobile_collapse=False, equal_height=True):
                         with gr.Row():
                             prompt = gr.Textbox(placeholder="Your prompt (what you want in place of what is erased)", show_label=False, elem_id="prompt")