app.py

import subprocess

import os
import sys
import subprocess

def run(cmd, cwd=None):
    print(f"▶ {cmd}")
    subprocess.check_call(cmd, shell=True, cwd=cwd)

def setup_deps():
    # Use a flag to prevent infinite restarts
    if os.environ.get("HF_SPACE_BOOTSTRAPPED") == "1":
        return

    # Try importing something to check if it's already set up
    try:
        import torch
        import sam2
        print("🔧 Dependencies already installed.")
        return  # all good, don't reinstall
    except ImportError:
        pass

    print("🔧 Installing dependencies...")
    run("pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu")
    run("pip install -e .", cwd="segment-anything-2")
    run("pip install --no-deps -r requirements_manual.txt")

    # Relaunch the script with an env flag to avoid looping
    print("♻️ Restarting app to apply changes...")
    os.environ["HF_SPACE_BOOTSTRAPPED"] = "1"
    os.execv(sys.executable, [sys.executable] + sys.argv)

setup_deps()

import gradio as gr
import numpy as np
from PIL import Image
import sam_utils
import matplotlib.pyplot as plt
from io import BytesIO

from sam2.sam2_image_predictor import SAM2ImagePredictor


# Dummy placeholders for SAM2 functions (replace with real logic)
def segment_reference(image, click):
    # click = [x, y]
    # Replace this with your SAM2 model's inference logic
    # Return a binary mask (numpy array with shape [H, W], values 0 or 1)
    print(f"Segmenting reference at point: {click}")
    width, height = image.size
    click = np.array(click)
    input_label = np.array([1 for _ in range(len(click))])
    sam2_img.set_image(image)

    masks, _, _ = sam2_img.predict(
        point_coords=click,
        point_labels=input_label,
        multimask_output=False,
    )

    return masks

def segment_target(target_images, ref_image, ref_mask):
    target_images = [np.array(target_image) for target_image in target_images]
    ref_image = np.array(ref_image)
    state = sam_utils.load_masks(sam2_vid, target_images, ref_image, ref_mask)
    out = sam_utils.propagate_masks(sam2_vid, state)[1:]
    return [mask['segmentation'] for mask in out]

def on_reference_upload(img):
    global click_coords
    click_coords = []  # clear the clicks
    return "Click Info: Cleared (new image uploaded)"

def visualize_segmentation(image, masks, target_images, target_masks):
    # Visualize the segmentation result
    num_tgt = len(target_images)
    fig, ax = plt.subplots(2, num_tgt, figsize=(6*num_tgt, 12))
    if num_tgt == 1:
        ax = np.expand_dims(ax, axis=1)
    ax[0][0].imshow(image.convert("L"), cmap='gray')
    for i, mask in enumerate(masks):
        sam_utils.show_mask(mask, ax[0][0], obj_id=i, alpha=0.75)
    ax[0][0].axis('off')
    ax[0][0].set_title("Reference Image with Expert Segmentation")
    for i in range(1, num_tgt):
        # set the rest to empty
        ax[0][i].axis('off')
    for i in range(num_tgt):
        ax[1][i].imshow(target_images[i].convert("L"), cmap='gray')
        for j, mask in enumerate(target_masks[i]):
            sam_utils.show_mask(mask, ax[1][i], obj_id=j, alpha=0.75)
        ax[1][i].axis('off')
        ax[1][i].set_title("Target Image with Inferred Segmentation")
    # save it to buffer
    plt.tight_layout()
    buf = BytesIO()
    plt.savefig(buf, format='png')
    buf.seek(0)
    vis = Image.open(buf).copy()
    plt.close(fig)
    buf.close()
    return vis

# Store click coords globally (can be improved with state)
click_coords = []

def record_click(img, evt: gr.SelectData):
    global click_coords
    click_coords.append([evt.index[0], evt.index[1]])
    return f"Clicked at: {click_coords}"

def generate(reference_image, target_images):
    global click_coords
    if not click_coords:
        return None, "Click on the reference image first!"
    
    target_images = [Image.open(f.name).convert("RGB").resize((1024,1024)) for f in target_images]

    ref_mask = segment_reference(reference_image, click_coords)
    tgt_masks = segment_target(target_images, reference_image, ref_mask)
    vis = visualize_segmentation(reference_image, ref_mask, target_images, tgt_masks)
    # clear the clicks
    click_coords = []
    return vis, "Done!"

with gr.Blocks() as demo:
    gr.Markdown("### SST Demo: Label-Efficient Trait Segmentation")
    
    with gr.Row():
        reference_img = gr.Image(type="pil", label="Reference Image")
        target_img = gr.File(file_types=["image"], file_count="multiple", label="Target Images")
    
    click_info = gr.Textbox(label="Click Info")
    generate_btn = gr.Button("Generate")
    output_mask = gr.Image(type="pil", label="Generated Mask")

    reference_img.select(fn=record_click, inputs=[reference_img], outputs=[click_info])
    reference_img.change(fn=on_reference_upload, inputs=[reference_img], outputs=[click_info])
    generate_btn.click(fn=generate, inputs=[reference_img, target_img], outputs=[output_mask, click_info])

global sam2_img
sam2_img = sam_utils.load_SAM2(ckpt_path="checkpoints/sam2_hiera_small.pt", model_cfg_path="checkpoints/sam2_hiera_s.yaml")
sam2_img = SAM2ImagePredictor(sam2_img)
global sam2_vid
sam2_vid = sam_utils.build_sam2_predictor(checkpoint="checkpoints/sam2_hiera_small.pt", model_cfg="checkpoints/sam2_hiera_s.yaml")
demo.launch()