test

app.py

@@ -10,7 +10,14 @@ def pil_to_torch(image):
     image = image.transpose((2, 0, 1))
     image = torch.tensor(image).float() / 255
     image = image.unsqueeze(0)
-    image = torch.nn.functional.interpolate(image, size=(128, 128*image.shape[3]//image.shape[2]))
+
+    ref_size = 256
+    if image.shape[2] > image.shape[3]:
+        size = (ref_size, ref_size * image.shape[2]//image.shape[3])
+    else:
+        size = (ref_size * image.shape[3]//image.shape[2], ref_size)
+
+    image = torch.nn.functional.interpolate(image, size=size, mode='bilinear')
     return image
 
 
@@ -27,7 +34,7 @@ def image_mod(image, noise_level, denoiser):
     if denoiser == 'DnCNN':
         denoiser = dinv.models.DnCNN()
     elif denoiser == 'MedianFilter':
-        denoiser = dinv.models.MedianFilter()
+        denoiser = dinv.models.MedianFilter(kernel_size=5)
     elif denoiser == 'BM3D':
         denoiser = dinv.models.BM3D()
     elif denoiser == 'TV':
@@ -58,7 +65,7 @@ demo = gr.Interface(
     examples=[['https://deepinv.github.io/deepinv/_static/deepinv_logolarge.png', 0.1, 'DnCNN']],
     outputs=[noise_image, output_images],
     title="Image Denoising with DeepInverse",
-    description="Denoise an image using a variety of denoisers and noise levels using the deepinverse library (https://deepinv.github.io/). We only include lightweight models like DnCNN and MedianFilter as this example is intended to be run on a CPU. We also automatically resize the input image to 128 vertical pixels to reduce the computation time. For more advanced models, please run the code locally.",
+    description="Denoise an image using a variety of denoisers and noise levels using the deepinverse library (https://deepinv.github.io/). We only include lightweight models like DnCNN and MedianFilter as this example is intended to be run on a CPU. We also automatically resize the input image to 256 vertical pixels to reduce the computation time. For more advanced models, please run the code locally.",
 )
 
 demo.launch()