Create app.py

app.py

@@ -0,0 +1,52 @@
+import torch
+import cv2
+    import numpy as np
+from PIL import Image
+import gradio as gr
+from segment_anything import SamPredictor, sam_model_registry
+from groundingdino.util.inference import load_model, predict, annotate
+
+grounding_dino_config = "groundingdino/config/GroundingDINO_SwinT_OGC.py"
+grounding_dino_weights = 'groundingdino_swift_ogc.path'
+
+dino_model = load_model(grounding_dino_config, grounding_dino_weights)
+
+sam_checkpoint = 'sam_vit_h_4b8939.pth'
+sam = sam_model_registry['vit_h'](checkpoint = sam_checkpoint)
+sam.to('cuda' if torch.cuda.is_available() else 'cpu')
+predictor = SamPredictor(sam)
+
+def grounded_sam_segment(image: Image.Image, prompt: str) -> Image.Image:
+    image_np = np.array(image.convert('RGB'))
+
+    boxes, logits, phrases = predict(
+        model = dino_model,
+        image = image_np,
+        caption = prompt,
+        box_threshold = 0.3,
+        text_threshold = 0.25
+    )
+
+    if len(boxes) == 0:
+        return image
+
+    predictor.set_image(image_np)
+    transformed_boxes = predictor.transform.apply_boxes_torch(boxes, image_np.shape[:2])
+    masks, _, = predictor.predict_torch(boxes = transformed_boxes, multimask_output=False)
+
+    mask = masks[0][0].cpu().numpy()
+    mask = np.stack([mask * 255] * 3, axis =-1).astype(np.units)
+    overlay = cv2.addweighted(image_np, 1, mask, 0.4, 0)
+    return Image.fromarray(overlay)
+
+    gr.Interface(
+        fn=grounded_sam_segment,
+        inputs=[
+            gr.Image(type='pil', label='Upload Image'),
+            gr.Textbox(label='Prompt', placeholder='e.g., cup handle, bottle')
+        ],
+        outputs=gr.Image(label='Segmented Output'),
+        title='Grounded-SAM Image Segmentation',
+        description="Accurate image segmentation using GroundingDINO + SAM. Prompt: 'cup handle', 'helmet', 'etc.'")
+        ]
+    ).launch()