input fix

app.py

@@ -20,8 +20,8 @@ device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
 image_size = 384
 transform = transforms.Compose([
-        transforms.Resize((image_size,image_size),interpolation=InterpolationMode.BICUBIC),
         transforms.ToTensor(),
+        transforms.Resize((image_size,image_size),interpolation=InterpolationMode.BICUBIC),
         transforms.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
         ]) 
 
@@ -31,6 +31,7 @@ model.eval()
 model = model.to(device)
 
 def inference(raw_image):
+    # raw_image = torch.tensor(raw_image)
     image = transform(raw_image).unsqueeze(0).to(device)
     with torch.no_grad():
           caption = model.generate(image, sample=False, num_beams=1, max_length=60, min_length=5) 
@@ -40,8 +41,16 @@ def inference(raw_image):
 inputs = [gr.Image(type='pil', interactive=False),]
 outputs = gr.outputs.Textbox(label="Caption")
 
-title = "FuseCap"
 description = "Gradio demo for FuseCap: Leveraging Large Language Models to Fuse Visual Data into Enriched Image Captions. This demo features a BLIP-based model, trained using FuseCap."
 
 article = "place holder"
-gr.Interface(inference, inputs, outputs, title=title, description=description, article=article, examples=[['birthday_dog.jpeg']]).launch(enable_queue=True)
+iface = gr.Interface(fn=inference, 
+                    inputs="image",
+                    outputs="text", 
+                    title="FuseCap",
+                    description=description,
+                    article=article,
+                    examples=[['birthday_dog.jpeg']],
+                    enable_queue=True)
+iface.launch()
+# gr.Interface(inference, inputs, outputs, title=title, description=description, article=article, examples=[['birthday_dog.jpeg']]).launch(enable_queue=True)