app.py

import gradio as gr
from transformers import SamModel, SamProcessor
from PIL import Image, ImageDraw
import torch
import numpy as np

# Load SAM
processor = SamProcessor.from_pretrained("facebook/sam-vit-huge")
model = SamModel.from_pretrained("facebook/sam-vit-huge")
model.eval()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Global variables
global_state = {
    "image": None,
    "clicks": [],  # List of tuples: (x, y, label) where label=1 fg, 0 bg
    "bbox": None,  # (x0, y0, x1, y1)
}

# Helper to apply mask overlay
def apply_mask_overlay(image: Image.Image, mask: np.ndarray, color=(255, 0, 0)) -> Image.Image:
    if mask.ndim == 3:
        mask = mask.squeeze()
    if mask.max() <= 1:
        mask = (mask * 255).astype(np.uint8)
    mask_img = Image.fromarray(mask).convert("L")
    color_mask = Image.new("RGB", image.size, color)
    blended = Image.composite(color_mask, image, mask_img)
    return Image.blend(image, blended, alpha=0.5)

# Set image
def upload_image(img):
    global_state["image"] = img
    global_state["clicks"] = []
    global_state["bbox"] = None
    return "Image uploaded. Now click or draw a box.", img

# Handle point clicks
def on_click(evt: gr.SelectData):
    if global_state["image"] is None:
        return gr.update(), "Please upload an image first."

    # Default to foreground click
    global_state["clicks"].append((evt.index[0], evt.index[1], 1))
    return gr.update(), f"Point added: ({evt.index[0]}, {evt.index[1]})"

# Handle bounding box input
def set_bbox(x0: int, y0: int, x1: int, y1: int):
    global_state["bbox"] = (x0, y0, x1, y1)
    return f"Bounding box set: ({x0}, {y0}, {x1}, {y1})"

# Run segmentation
def run_segmentation():
    if global_state["image"] is None:
        return None, "Please upload an image first."

    image = global_state["image"]
    inputs = processor(image, return_tensors="pt").to(device)

    if global_state["clicks"]:
        coords = [[(x, y) for (x, y, l) in global_state["clicks"]]]
        labels = [[l for (_, _, l) in global_state["clicks"]]]
        input_points = torch.tensor([coords], device=device)  # shape [1, 1, N, 2]
        input_labels = torch.tensor([labels], device=device)  # shape [1, 1, N]

        inputs.update({
            "input_points": input_points,
            "input_labels": input_labels
            })

    if global_state["bbox"]:
        x0, y0, x1, y1 = global_state["bbox"]
        box = torch.tensor([[[x0, y0, x1, y1]]], device=device)
        inputs.update({"input_boxes": box})

    with torch.no_grad():
        outputs = model(**inputs, multimask_output=False)

    masks = processor.image_processor.post_process_masks(
        outputs.pred_masks.cpu(),
        inputs["original_sizes"].cpu(),
        inputs["reshaped_input_sizes"].cpu()
    )[0]

    final_mask = masks[0].numpy().astype(np.uint8)  # shape: (H, W)
    overlayed = apply_mask_overlay(image.convert("RGB"), final_mask)

    return overlayed, "Segmentation complete."

# Reset
def reset_all():
    global_state["image"] = None
    global_state["clicks"] = []
    global_state["bbox"] = None
    return None, None, "State reset."

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("Interactive Pathology Segmentation with SAM")
    
    with gr.Row():
        with gr.Column():
            image_input = gr.Image(type="pil", label="Upload Pathology Image")
            upload_status = gr.Textbox(label="Status")
            upload_btn = gr.Button("Upload & Set Image")
            click_output = gr.Textbox(label="Click Info")
            run_btn = gr.Button("Run Segmentation")
            reset_btn = gr.Button("Reset")
            bbox_coords = gr.Textbox(label="Manual BBox (x0,y0,x1,y1)")
            set_bbox_btn = gr.Button("Set Bounding Box")
            examples = gr.Examples(
                examples=[
                    ["images/image.webp"],
                    ["images/image2.webp"],
                    ["images/image3.webp"]
                ],
                inputs=[image_input],
                label="Example Pathology Images"
            )
        with gr.Column():
            image_output = gr.Image(type="pil", label="Segmentation Output")

    # Handlers
    upload_btn.click(upload_image, inputs=image_input, outputs=[upload_status, image_output])
    image_input.select(on_click, outputs=[image_output, click_output])
    run_btn.click(run_segmentation, outputs=[image_output, upload_status])
    reset_btn.click(reset_all, outputs=[image_output, click_output, upload_status])
    set_bbox_btn.click(
        lambda coords: set_bbox(*map(int, coords.split(","))),
        inputs=bbox_coords,
        outputs=click_output
    )

demo.launch()