Create app.py

app.py

@@ -0,0 +1,68 @@
+import gradio as gr
+from PIL import Image, ImageChops
+from transformers import BlipProcessor, BlipForQuestionAnswering
+import torch
+
+# Load BLIP model
+model_name = "Salesforce/blip-vqa-base"
+processor = BlipProcessor.from_pretrained(model_name)
+model = BlipForQuestionAnswering.from_pretrained(model_name)
+
+# Ensure device
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model.to(device)
+
+def crop_difference(base_img: Image.Image, trash_img: Image.Image) -> Image.Image:
+    # Convert to same mode
+    base_img = base_img.convert("RGB")
+    trash_img = trash_img.convert("RGB")
+    
+    # Compute difference
+    diff = ImageChops.difference(trash_img, base_img)
+    # Crop to non-zero bbox
+    bbox = diff.getbbox()
+    if bbox:
+        cropped = trash_img.crop(bbox)
+        return cropped
+    else:
+        return trash_img  # fallback if no difference
+
+def identify_material(base_img: Image.Image, trash_img: Image.Image) -> str:
+    if base_img is None or trash_img is None:
+        return "Please upload both base and trash images."
+    
+    cropped = crop_difference(base_img, trash_img)
+    
+    question = "What material is this? Choose from: plastic, metal, paper, cardboard, glass, trash."
+    inputs = processor(cropped, question, return_tensors="pt").to(device)
+    out = model.generate(**inputs)
+    answer = processor.decode(out[0], skip_special_tokens=True)
+    
+    valid_classes = ["plastic", "metal", "paper", "cardboard", "glass", "trash"]
+    result = next((c for c in valid_classes if c in answer.lower()), "trash")
+    return result.capitalize()
+
+title = "Smart Waste Material Detector"
+description = """
+Upload two images:
+1. Base image (empty background)
+2. Trash image (object placed on background)
+
+The AI will detect the difference and classify the material:
+plastic, metal, paper, cardboard, glass, or trash.
+"""
+
+demo = gr.Interface(
+    fn=identify_material,
+    inputs=[
+        gr.Image(type="pil", label="Upload Base Image (Empty)"),
+        gr.Image(type="pil", label="Upload Trash Image")
+    ],
+    outputs=gr.Textbox(label="Detected Material"),
+    title=title,
+    description=description,
+    allow_flagging="never"
+)
+
+if __name__ == "__main__":
+    demo.launch()