Create app.py

app.py

@@ -0,0 +1,27 @@
+import gradio as gr
+import torch
+import torchvision.transforms as transforms
+from PIL import Image
+
+def predict(image):
+    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+    model = torchvision.models.resnet50(pretrained=True).to(device)
+    model.eval()
+    transform = transforms.Compose([
+        transforms.Resize(256),
+        transforms.CenterCrop(224),
+        transforms.ToTensor(),
+        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+    ])
+
+    img = Image.fromarray(image.astype('uint8'), 'RGB')
+    img = transform(img).unsqueeze(0).to(device)
+    with torch.no_grad():
+        output = model(img)
+    _, predicted = output.max(1)
+    return predicted.item()
+
+input_image = gr.inputs.Image()
+output_text = gr.outputs.Textbox()
+
+gr.Interface(fn=predict, inputs=input_image, outputs=output_text).launch()