Create app.py

app.py

@@ -0,0 +1,59 @@
+import gradio as gr
+from transformers import AutoProcessor, AutoModelForVision2Seq
+import torch
+from PIL import Image
+
+# Load model and processor
+MODEL_ID = "liuhaotian/llava-v1.6-vicuna-7b"
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+model = AutoModelForVision2Seq.from_pretrained(MODEL_ID, torch_dtype=torch.float16, low_cpu_mem_usage=True)
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model.to(device)
+
+base_image = None
+
+def set_base(image):
+    global base_image
+    base_image = image
+    return "✅ Base image has been set."
+
+def detect_object(image, prompt):
+    if base_image is None:
+        return "⚠️ Please upload a base image first."
+
+    query = (
+        f"Ignore the base image and focus only on new or added objects. "
+        f"Base image and detection image are given. {prompt or 'Describe and identify the materials of new objects.'}"
+    )
+
+    inputs = processor(
+        text=query,
+        images=[base_image, image],
+        return_tensors="pt"
+    ).to(device, torch.float16)
+
+    output = model.generate(**inputs, max_new_tokens=256)
+    result = processor.decode(output[0], skip_special_tokens=True)
+    return result
+
+# Build Gradio UI
+with gr.Blocks(title="LLaVA Object & Material Detector") as demo:
+    gr.Markdown("## 🧠 LLaVA-1.6 (Vicuna-7B) Object & Material Detection\nUpload a base image first, then upload another image to detect new objects while ignoring the base.")
+
+    with gr.Row():
+        with gr.Column():
+            base_img = gr.Image(label="Base Image", type="pil")
+            set_base_btn = gr.Button("Set as Base Image")
+            set_base_status = gr.Textbox(label="Status")
+
+        with gr.Column():
+            target_img = gr.Image(label="Detection Image", type="pil")
+            user_prompt = gr.Textbox(label="Custom Instruction", placeholder="e.g. Detect added objects and describe their material.")
+            run_btn = gr.Button("Run Detection")
+            output_box = gr.Textbox(label="Model Output")
+
+    set_base_btn.click(set_base, inputs=base_img, outputs=set_base_status)
+    run_btn.click(detect_object, inputs=[target_img, user_prompt], outputs=output_box)
+
+demo.launch()