app.py

import gradio as gr
import requests
import base64
import numpy as np
import cv2

# Hugging Face SAM API 配置
HF_API_URL = "https://api-inference.huggingface.co/models/facebook/sam-vit-base"
HF_TOKEN = "YOUR_HUGGINGFACE_TOKEN"  # 替换成你自己的 token
headers = {"Authorization": f"Bearer {HF_TOKEN}"}


def segment_and_crop(image):
    # 将输入图像转为 PNG bytes
    _, buffer = cv2.imencode(".png", cv2.cvtColor(image, cv2.COLOR_RGB2BGR))
    img_bytes = buffer.tobytes()

    # 调用 Hugging Face SAM API
    response = requests.post(HF_API_URL, headers=headers, files={"image": img_bytes})
    if response.status_code != 200:
        return None, f"API error: {response.status_code}\n{response.text}"

    result = response.json()

    # SAM 输出 mask（有的模型返回 'masks' base64）
    masks = result.get("masks", [])
    if not masks:
        return None, "No masks returned"

    cropped_images = []
    for idx, mask_b64 in enumerate(masks):
        mask_bytes = base64.b64decode(mask_b64)
        mask_array = cv2.imdecode(np.frombuffer(mask_bytes, np.uint8), cv2.IMREAD_GRAYSCALE)
        ys, xs = np.where(mask_array > 127)
        if len(xs) == 0 or len(ys) == 0:
            continue

        # 取边界框并裁剪
        x1, x2 = xs.min(), xs.max()
        y1, y2 = ys.min(), ys.max()
        cropped = image[y1:y2, x1:x2]
        cropped_images.append(cropped)

    if not cropped_images:
        return None, "No valid crops"

    # 输出多张裁剪图
    return cropped_images, f"Detected {len(cropped_images)} object(s)."


demo = gr.Interface(
    fn=segment_and_crop,
    inputs=gr.Image(type="numpy", label="Upload or Capture an Image"),
    outputs=[gr.Gallery(label="Cropped Objects"), gr.Textbox(label="Status")],
    title="Smart Object Cropper with SAM",
    description="Upload or capture an image. The app will use the Segment Anything Model (SAM) on Hugging Face to detect and crop main objects.",
)

if __name__ == "__main__":
    demo.launch()