app.py

import spaces
import gradio as gr
import torch
import numpy as np
from PIL import Image
import matplotlib
from transformers import Sam3Processor, Sam3Model
import warnings
warnings.filterwarnings("ignore")

# Global model and processor
device = "cuda" if torch.cuda.is_available() else "cpu"
model = Sam3Model.from_pretrained("facebook/sam3", torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32).to(device)
processor = Sam3Processor.from_pretrained("facebook/sam3")

def overlay_masks(image: Image.Image, masks: torch.Tensor) -> Image.Image:
    """
    Overlay segmentation masks on the input image using rainbow colormap.
    """
    image = image.convert("RGBA")
    masks = 255 * masks.cpu().numpy().astype(np.uint8)
    
    n_masks = masks.shape[0]
    if n_masks == 0:
        return image.convert("RGB")
    
    cmap = matplotlib.colormaps.get_cmap("rainbow").resampled(n_masks)
    colors = [
        tuple(int(c * 255) for c in cmap(i)[:3])
        for i in range(n_masks)
    ]

    for mask, color in zip(masks, colors):
        mask_img = Image.fromarray(mask)
        overlay = Image.new("RGBA", image.size, color + (0,))
        alpha = mask_img.point(lambda v: int(v * 0.5))
        overlay.putalpha(alpha)
        image = Image.alpha_composite(image, overlay)
    return image

spaces.GPU()

def segment(image: Image.Image, text: str, threshold: float, mask_threshold: float):
    """
    Perform promptable concept segmentation using SAM3.
    """
    if image is None:
        return None, "❌ Please upload an image."
    
    try:
        inputs = processor(images=image, text=text.strip(), return_tensors="pt").to(device)
        
        with torch.no_grad():
            outputs = model(**inputs)
        
        results = processor.post_process_instance_segmentation(
            outputs,
            threshold=threshold,
            mask_threshold=mask_threshold,
            target_sizes=inputs.get("original_sizes").tolist()
        )[0]
        
        n_masks = len(results['masks'])
        if n_masks == 0:
            return image, f"❌ No objects found matching '{text}' (try adjusting thresholds or changing prompt)."
        
        overlaid_image = overlay_masks(image, results["masks"])
        
        scores_text = ", ".join([f"{s:.2f}" for s in results['scores'].cpu().numpy()[:5]])  # Top 5 scores
        info = f"✅ Found **{n_masks}** objects matching **'{text}'**\nConfidence scores: {scores_text}{'...' if n_masks > 5 else ''}"
        
        return overlaid_image, info
        
    except Exception as e:
        return image, f"❌ Error during segmentation: {str(e)}"

def clear_all():
    """Clear all inputs and outputs"""
    return None, "", None, 0.5, 0.5

def segment_example(image_path: str, prompt: str):
    """Handle example clicks"""
    image = Image.open(image_path) if image_path else None
    return segment(image, prompt, 0.5, 0.5)

# Gradio Interface
with gr.Blocks(
    theme=gr.themes.Soft(),
    title="SAM3 - Promptable Concept Segmentation",
    css="""
    .gradio-container {max-width: 1400px !important;}
    """
) as demo:
    gr.Markdown(
        """
        # SAM3 - Promptable Concept Segmentation (PCS)
        
        **SAM3** performs zero-shot instance segmentation using natural language prompts on images.
        Upload an image, enter a text prompt (e.g., "person", "car", "dog"), and get segmentation masks for all matching objects.
        
        Built with [anycoder](https://huggingface.co/spaces/akhaliq/anycoder)
        """
    )
    
    gr.Markdown("### Inputs")
    with gr.Row(variant="panel"):
        image_input = gr.Image(
            label="Input Image",
            type="pil",
            height=400,
        )
        image_output = gr.Image(
            label="Output (Segmented Image)",
            height=400,
            interactive=False
        )
    
    with gr.Row():
        text_input = gr.Textbox(
            label="Text Prompt",
            placeholder="e.g., a person, ear, cat, bicycle...",
            scale=3
        )
        clear_btn = gr.Button("🔍 Clear", size="sm", variant="secondary")
    
    with gr.Row():
        thresh_slider = gr.Slider(
            minimum=0.0,
            maximum=1.0,
            value=0.5,
            step=0.01,
            label="Detection Threshold",
            info="Higher values = fewer detections (objectness confidence)"
        )
        mask_thresh_slider = gr.Slider(
            minimum=0.0,
            maximum=1.0,
            value=0.5,
            step=0.01,
            label="Mask Threshold",
            info="Higher values = sharper masks"
        )
    
    info_output = gr.Markdown(
        value="📝 Enter a prompt and click **Segment** to start.",
        label="Info / Results"
    )
    
    segment_btn = gr.Button("🎯 Segment", variant="primary", size="lg")
    
    # Clear button handler
    clear_btn.click(
        fn=clear_all,
        outputs=[image_input, text_input, image_output, thresh_slider, mask_thresh_slider]
    )
    
    # Segment button handler
    segment_btn.click(
        fn=segment,
        inputs=[image_input, text_input, thresh_slider, mask_thresh_slider],
        outputs=[image_output, info_output]
    ).then(
        fn=lambda: None,
    )
    
    # Examples
    gr.Markdown("### Examples")
    examples = [
        [
            "http://images.cocodataset.org/val2017/000000077595.jpg",
            "ear"
        ],
        [
            "http://images.cocodataset.org/val2017/000000039769.jpg",
            "cat"
        ],
        [
            "http://images.cocodataset.org/val2017/000000001247.jpg",
            "person"
        ],
        [
            "http://images.cocodataset.org/val2017/000000521315.jpg",
            "bicycle"
        ],
        [
            "http://images.cocodataset.org/val2017/000000029369.jpg",
            "dog"
        ]
    ]
    gr.Examples(
        examples=examples,
        inputs=[image_input, text_input],
        fn=segment_example,
        outputs=[image_output, info_output],
        cache_examples=True,
        examples_per_page=10,
        label="Try these COCO examples (URLs auto-load)"
    )
    
    gr.Markdown(
        """
        ### Notes
        - **Model**: [facebook/sam3](https://huggingface.co/facebook/sam3)
        - Supports natural language prompts like "a red car" or simple nouns.
        - GPU recommended for faster inference.
        - Thresholds control detection sensitivity and mask quality.
        """
    )

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860, share=False, debug=True)
```

=== utils.py ===
```python
import torch
import numpy as np
from PIL import Image
import matplotlib
import requests
from io import BytesIO

def load_image_from_url(url: str) -> Image.Image:
    """
    Load an image from a URL.
    
    Args:
        url: Image URL
        
    Returns:
        PIL Image object
    """
    try:
        response = requests.get(url, timeout=10)
        response.raise_for_status()
        image = Image.open(BytesIO(response.content))
        return image.convert("RGB")
    except Exception as e:
        raise ValueError(f"Could not load image from URL: {str(e)}")

def validate_image(image: Image.Image) -> bool:
    """
    Validate if the image is suitable for processing.
    
    Args:
        image: PIL Image object
        
    Returns:
        True if valid, False otherwise
    """
    if image is None:
        return False
    
    if image.size[0] <= 0 or image.size[1] <= 0:
        return False
        
    return True

def resize_for_processing(image: Image.Image, max_size: int = 1024) -> Image.Image:
    """
    Resize image for processing while maintaining aspect ratio.
    
    Args:
        image: Input PIL Image
        max_size: Maximum size for the longer dimension
        
    Returns:
        Resized PIL Image
    """
    width, height = image.size
    if max(width, height) <= max_size:
        return image
    
    if width > height:
        new_width = max_size
        new_height = int(height * max_size / width)
    else:
        new_height = max_size
        new_width = int(width * max_size / height)
    
    return image.resize((new_width, new_height), Image.Resampling.LANCZOS)

def overlay_masks_advanced(image: Image.Image, masks: torch.Tensor, alpha: float = 0.5) -> Image.Image:
    """
    Advanced overlay function with customizable alpha.
    
    Args:
        image: Input PIL Image
        masks: Segmentation masks tensor
        alpha: Overlay transparency (0-1)
        
    Returns:
        Overlaid PIL Image
    """
    image = image.convert("RGBA")
    masks = 255 * masks.cpu().numpy().astype(np.uint8)
    
    n_masks = masks.shape[0]
    if n_masks == 0:
        return image.convert("RGB")
    
    # Use a good colormap
    cmap = matplotlib.colormaps.get_cmap("tab10").resampled(n_masks)
    colors = [
        tuple(int(c * 255) for c in cmap(i)[:3])
        for i in range(n_masks)
    ]

    for mask, color in zip(masks, colors):
        mask_img = Image.fromarray(mask)
        overlay = Image.new("RGBA", image.size, color + (0,))
        alpha_map = mask_img.point(lambda v: int(v * alpha * 255))
        overlay.putalpha(alpha_map)
        image = Image.alpha_composite(image, overlay)
    
    return image