sam2_mask.py

import spaces
import gradio as gr
import os
os.environ["TORCH_CUDNN_SDPA_ENABLED"] = "1"
import torch
import numpy as np
import cv2
import matplotlib.pyplot as plt
from PIL import Image, ImageFilter
from sam2.build_sam import build_sam2
from sam2.sam2_image_predictor import SAM2ImagePredictor

def preprocess_image(image):
    return image, gr.State([]), gr.State([]), image

def get_point(point_type, tracking_points, trackings_input_label, original_image, evt):
    x, y = evt.index
    tracking_points.append((x, y))
    trackings_input_label.append(1 if point_type == "include" else 0)
    
    # Redraw all points on original image
    w, h = original_image.size
    radius = int(min(w, h) * 0.02)
    img = original_image.convert("RGBA")
    draw = ImageDraw.Draw(img)
    for i, (cx, cy) in enumerate(tracking_points):
        color = (0, 255, 0, 255) if trackings_input_label[i] == 1 else (255, 0, 0, 255)
        draw.ellipse([cx-radius, cy-radius, cx+radius, cy+radius], fill=color)
    return tracking_points, trackings_input_label, img

# use bfloat16 for the entire notebook
torch.autocast(device_type="cuda", dtype=torch.bfloat16).__enter__()
if torch.cuda.get_device_properties(0).major >= 8:
    # turn on tfloat32 for Ampere GPUs (https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices)
    torch.backends.cuda.matmul.allow_tf32 = True
    torch.backends.cudnn.allow_tf32 = True

def show_mask(mask, ax, random_color=False, borders=True):
    if random_color:
        color = np.concatenate([np.random.random(3), np.array([0.6])], axis=0)
    else:
        color = np.array([30/255, 144/255, 255/255, 0.6])
    h, w = mask.shape[-2:]
    mask = mask.astype(np.uint8)
    mask_image = mask.reshape(h, w, 1) * color.reshape(1, 1, -1)
    if borders:
        import cv2
        contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
        # Try to smooth contours
        contours = [cv2.approxPolyDP(contour, epsilon=0.01, closed=True) for contour in contours]
        mask_image = cv2.drawContours(mask_image, contours, -1, (1, 1, 1, 0.5), thickness=2)
    ax.imshow(mask_image)

def show_points(coords, labels, ax, marker_size=375):
    pos_points = coords[labels == 1]
    neg_points = coords[labels == 0]
    ax.scatter(pos_points[:, 0], pos_points[:, 1], color='green', marker='*', s=marker_size, edgecolor='white', linewidth=1.25)
    ax.scatter(neg_points[:, 0], neg_points[:, 1], color='red', marker='*', s=marker_size, edgecolor='white', linewidth=1.25)

def show_box(box, ax):
    x0, y0 = box[0], box[1]
    w, h = box[2] - box[0], box[3] - box[1]
    ax.add_patch(plt.Rectangle((x0, y0), w, h, edgecolor='green', facecolor=(0, 0, 0, 0), lw=2))

def show_masks(image, masks, scores, point_coords=None, box_coords=None, input_labels=None, borders=True):
    combined_images = []  # List to store filenames of images with masks overlaid
    mask_images = []      # List to store filenames of separate mask images
    for i, (mask, score) in enumerate(zip(masks, scores)):
        # ---- Original Image with Mask Overlaid ----
        plt.figure(figsize=(10, 10))
        plt.imshow(image)
        show_mask(mask, plt.gca(), borders=borders)  # Draw the mask with borders
        if box_coords is not None:
            show_box(box_coords, plt.gca())
        if len(scores) > 1:
            plt.title(f"Mask {i+1}, Score: {score:.3f}", fontsize=18)
        plt.axis('off')
        # Save the figure as a JPG file
        combined_filename = f"combined_image_{i+1}.jpg"
        plt.savefig(combined_filename, format='jpg', bbox_inches='tight')
        combined_images.append(combined_filename)
        plt.close()  # Close the figure to free up memory
        # ---- Separate Mask Image (White Mask on Black Background) ----
        # Create a black image
        mask_image = np.zeros_like(image, dtype=np.uint8)
        # The mask is a binary array where the masked area is 1, else 0.
        # Convert the mask to a white color in the mask_image
        mask_layer = (mask > 0).astype(np.uint8) * 255
        for c in range(3):  # Assuming RGB, repeat mask for all channels
            mask_image[:, :, c] = mask_layer
        # Save the mask image
        mask_filename = f"mask_image_{i+1}.png"
        Image.fromarray(mask_image).save(mask_filename)
        mask_images.append(mask_filename)
        plt.close()  # Close the figure to free up memory
    return combined_images, mask_images

@spaces.GPU()
def sam_process(original_image, points, labels):
    # Convert image to numpy array for SAM2 processing
    image = np.array(original_image)
    predictor = SAM2ImagePredictor.from_pretrained("facebook/sam2.1-hiera-large")
    predictor.set_image(image)
    input_point = np.array(points)
    input_label = np.array(labels)
    masks, scores, _ = predictor.predict(input_point, input_label, multimask_output=False)
    sorted_indices = np.argsort(scores)[::-1]
    masks = masks[sorted_indices]
    
    # Generate mask image
    mask = masks[0] * 255
    mask_image = Image.fromarray(mask.astype(np.uint8))
    return mask_image
    # sam2_model = build_sam2(model_cfg, sam2_checkpoint, device="cuda")
    # predictor = SAM2ImagePredictor(sam2_model)


def create_sam2_tab():
    first_frame = gr.State()  # Tracks original image
    tracking_points = gr.State([])
    trackings_input_label = gr.State([])
    
    with gr.Column():
        gr.Markdown("# SAM2 Image Predictor")
        gr.Markdown("1. Upload your image\n2. Click points to mask\n3. Submit")
        points_map = gr.Image(label="Points Map", type="pil", interactive=True)
        input_image = gr.Image(type="pil", visible=False)  # Original image
        
        with gr.Row():
            point_type = gr.Radio(["include", "exclude"], value="include", label="Point Type")
            clear_button = gr.Button("Clear Points")
        submit_button = gr.Button("Submit")
        output_image = gr.Image("Segmented Output")
        
        # Event handlers
        points_map.upload(
            lambda img: (img, img, [], []),
            inputs=points_map,
            outputs=[input_image, first_frame, tracking_points, trackings_input_label]
        )
        clear_button.click(
            lambda img: ([], [], img),
            inputs=first_frame,
            outputs=[tracking_points, trackings_input_label, points_map]
        )
        points_map.select(
            get_point,
            inputs=[point_type, tracking_points, trackings_input_label, first_frame],
            outputs=[tracking_points, trackings_input_label, points_map]
        )
        submit_button.click(
            sam_process,
            inputs=[input_image, tracking_points, trackings_input_label],
            outputs=output_image
        )
    
    return input_image, points_map, output_image