Fix resolution

app.py

@@ -67,7 +67,7 @@ def greet(image, audio):
     seg_image = ((1 - seg.squeeze().detach().cpu().numpy()) * 255).astype(np.uint8)
     seg_image = Image.fromarray(seg_image)
     heatmap_image = cv2.applyColorMap(np.array(seg_image), cv2.COLORMAP_JET)
-    overlaid_image = cv2.addWeighted(np.array(image), 0.5, heatmap_image, 0.5, 0)
+    overlaid_image = cv2.addWeighted(np.array(image.resize(352, 352)), 0.5, heatmap_image, 0.5, 0)
 
     return overlaid_image
 
@@ -75,8 +75,9 @@ def greet(image, audio):
 title = "Audio-Grounded Contrastive Learning"
 
 description = """<p>
-This is a simple demo of Can CLIP Help Sound Source Localization? (WACV2024), zero-shot visual sound localization.<br><br>
-To use it simply upload an image and corresponding audio to mask (identify in the image), or use one of the examples below and click ‘submit’.
+This is a simple demo of our WACV'24 paper 'Can CLIP Help Sound Source Localization?', zero-shot visual sound localization.<br><br>
+To use it simply upload an image and corresponding audio to mask (identify in the image), or use one of the examples below and click ‘submit’.<br><br>
+Results will show up in a few seconds.
 </p>"""
 
 article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2311.04066'>Can CLIP Help Sound Source Localization?</a> | <a href='https://github.com/swimmiing/ACL-SSL'>Offical Github repo</a></p>"