input scale selectable

app.py

@@ -11,6 +11,7 @@ from models.modelNetB import Generator as GB
 from models.modelNetC import Generator as GC
 
 scale_size = 128
+scale_sizes = [128, 256, 512]
 # load model
 modeltype2path = {
     'ModelA': 'DTM_exp_train10%_model_a/g-best.pth',
@@ -33,8 +34,9 @@ preprocess = transforms.Compose([
     transforms.ToTensor()
 ])
 
-def predict(input_image, model_name):
+def predict(input_image, model_name, input_scale_factor):
     pil_image = Image.fromarray(input_image.astype('uint8'), 'RGB')
+    pil_image = transforms.Resize((input_scale_factor, input_scale_factor))(pil_image)
     # transform image to torch and do preprocessing
     torch_img = preprocess(pil_image).to(DEVICE).unsqueeze(0).to(DEVICE)
     torch_img = (torch_img - torch.min(torch_img)) / (torch.max(torch_img) - torch.min(torch_img))
@@ -61,8 +63,9 @@ def predict(input_image, model_name):
 iface = gr.Interface(
     fn=predict, 
     inputs=[
-        gr.Image(shape=(scale_size,scale_size)),
-        gr.inputs.Radio(MODELS_TYPE)
+        gr.Image(),
+        gr.inputs.Radio(MODELS_TYPE),
+        gr.inputs.Radio(scale_sizes)
     ], 
     outputs=[
         gr.Text(label='Model info'),
@@ -70,9 +73,9 @@ iface = gr.Interface(
         gr.Image(label='DTM')
     ],
     examples=[
-        [f"demo_imgs/{name}", MODELS_TYPE[0]] for name in os.listdir('demo_imgs')
+        [f"demo_imgs/{name}", MODELS_TYPE[0], 128] for name in os.listdir('demo_imgs')
     ],
     title="Super Resolution and DTM Estimation",
-description=f"This demo predict Super Resolution and (Super Resolution) DTM from a Grayscale image (if RGB we convert it, for demo reason input is scale to {scale_size}x{scale_size})."
+    description=f"This demo predict Super Resolution and (Super Resolution) DTM from a Grayscale image (if RGB we convert it)."
 )
 iface.launch()