fffiloni commited on
Commit
485b322
·
verified ·
1 Parent(s): 970b009

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +17 -4
app.py CHANGED
@@ -159,11 +159,24 @@ def infer_inp(prompt, audio_path, spec_with_mask, progress=gr.Progress(track_tqd
159
  raw_image = image_add_color(torch_to_pil(norm_spec))
160
 
161
  # Add Mask
162
- mask = torch.zeros_like(norm_spec)[:1,...]
163
- mask[:, :, width_start:width_start+width] = 1
164
- mask_image = torch_to_pil(mask)
 
 
 
 
 
 
 
 
 
 
 
 
 
165
 
166
- mask, masked_spec = prepare_mask_and_masked_image(norm_spec, mask)
167
  masked_spec_image = torch_to_pil(masked_spec)
168
 
169
  # color masked spec and paint masked area to black
 
159
  raw_image = image_add_color(torch_to_pil(norm_spec))
160
 
161
  # Add Mask
162
+ #mask = torch.zeros_like(norm_spec)[:1,...]
163
+ #mask[:, :, width_start:width_start+width] = 1
164
+ #mask_image = torch_to_pil(mask)
165
+
166
+ # Load the mask image (input from user)
167
+ mask_pil = spec_with_mask['layers'][0]
168
+
169
+ # Convert to tensor and normalize
170
+ mask_tensor = transforms.ToTensor()(mask_pil) # Shape: (1, H, W), values in [0, 1]
171
+
172
+ # Ensure the shape matches expected input (add batch dimension if needed)
173
+ mask_tensor = mask_tensor[:1, :, :] # Keep only one channel (grayscale)
174
+ mask_tensor = mask_tensor.to(device, dtype) # Send to correct device and dtype
175
+
176
+ # Convert to PIL image if needed for visualization
177
+ mask_image = torch_to_pil(mask_tensor)
178
 
179
+ mask, masked_spec = prepare_mask_and_masked_image(norm_spec, mask_tensor)
180
  masked_spec_image = torch_to_pil(masked_spec)
181
 
182
  # color masked spec and paint masked area to black