Update app.py

app.py

@@ -10,7 +10,7 @@ import io
 import gradio as gr
 import PIL
 
-model_choice = 2
+model_choice = 3
 model_names = ["nvidia/mit-b0",'facebook/convnext-base-224', 'microsoft/resnet-18', 'microsoft/swin-tiny-patch4-window7-224']
 model_name = model_names[model_choice]
 device     = 'cuda' if torch.cuda.is_available() else 'cpu'
@@ -137,7 +137,7 @@ def saliency_graph(img1, steps=25):
 
 # gradio Interface
 def gradio_interface(img):
-    smoothgrad_mask_grayscale, fig_img = saliency_graph(img, steps=12)
+    smoothgrad_mask_grayscale, fig_img = saliency_graph(img, steps=20)
     return smoothgrad_mask_grayscale, fig_img
 
 with gr.Blocks(title='Looking at the pixels models attend to', description="This function finds the most critical pixels in an image for predicting a class. The best models will ideally make predictions by highlighting the expected object. Poorly generalizable models will often rely on environmental cues instead and forego looking at the most important pixels. Highlighting the most important pixels helps explain/build trust about whether a given model uses the correct features to make its prediction.", live=True) as iface:
@@ -147,7 +147,7 @@ with gr.Blocks(title='Looking at the pixels models attend to', description="This
         with gr.Column():
             test_image = gr.Image(label="Input Image", live=True)
             input_btn = gr.Button(label="Classify image")
-            model_select_dropdown = gr.Radio(model_names, label="Model to test", interactive=True, default=2)
+            model_select_dropdown = gr.Radio(model_names, label="Model to test", interactive=True, default=3)
         with gr.Column():
             output = gr.Image(label="Pixels used for classification")
             output2 = gr.Image(label="Top 5 Predictions")