app.py

import gradio as gr
from gradio_bbox_annotator import BBoxAnnotator
from PIL import Image
import numpy as np
import torch
import os
from inference import load_model, run  # 注意是你新改过的 inference.py
import shutil
import os
import os
import subprocess

# 示例：查看当前目录大小
print("===== Space 中的目录结构和占用 =====")
subprocess.run("du -sh *", shell=True)

# 示例：列出缓存目录
print("===== .cache 目录 =====")
subprocess.run("ls -lh ~/.cache", shell=True)
# 删除 Hugging Face 默认缓存目录（~/.cache）
cache_path = os.path.expanduser("~/.cache")
if os.path.exists(cache_path):
    shutil.rmtree(cache_path)
    print("✅ Deleted ~/.cache to free space.")
# 初始化模型全局变量
MODEL = None
DEVICE = torch.device("cpu")
CUDA_READY = False

# 加载模型（默认 CPU）
def load_model_cpu():
    global MODEL, DEVICE
    MODEL, DEVICE = load_model()  # 你新写的load_model()封装在 inference.py 里

load_model_cpu()

# 自动 GPU 切换（Spaces 环境支持）
def prepare_cuda():
    global MODEL, DEVICE, CUDA_READY
    if torch.cuda.is_available() and not CUDA_READY:
        print("CUDA is available. Moving model to GPU...")
        MODEL.to("cuda")
        DEVICE = torch.device("cuda")
        CUDA_READY = True
        _ = torch.zeros(1, device=DEVICE)
        print("Model moved to CUDA.")
    else:
        print("CUDA not available or already initialized.")

# 解析框坐标
def parse_first_bbox(bboxes):
    if not bboxes:
        return None
    b = bboxes[0]
    if isinstance(b, dict):
        x, y = float(b["x"]), float(b["y"])
        w, h = float(b["width"]), float(b["height"])
        return x, y, x + w, y + h
    if isinstance(b, (list, tuple)) and len(b) >= 4:
        return float(b[0]), float(b[1]), float(b[2]), float(b[3])
    return None

# 推理逻辑：输入图像和 bbox，输出 mask + 坐标信息
def segment(annot_value):
    prepare_cuda()  # 如果可用 GPU 就切换

    if annot_value is None or len(annot_value) < 1:
        return None, 

    img_path = annot_value[0]
    bboxes = annot_value[1] if len(annot_value) > 1 else []

    if not bboxes:
        return None, 

    box = parse_first_bbox(bboxes)
    if box is None:
        return None, "解析矩形框失败，请重画。"

    xmin, ymin, xmax, ymax = map(int, [box[0], box[1], box[2], box[3]])
    box_array = [[xmin, ymin, xmax, ymax]]

    # 推理
    mask = run(MODEL, img_path, box_array, DEVICE)

    mask_rgb = np.stack([mask * 255] * 3, axis=-1).astype(np.uint8)
    bbox_text = f"xmin={xmin}, ymin={ymin}, xmax={xmax}, ymax={ymax}"

    return Image.fromarray(mask_rgb), bbox_text

# 示例图（可选）
example = ("003_img.png", [(50, 60, 120, 150, "cell")])

# 创建 Gradio 界面
demo = gr.Interface(
    fn=segment,
    inputs=BBoxAnnotator(
        value=example,
        categories=["cell", "nucleus"],
        label="Upload Microscopy Image"
    ),
    outputs=[
        gr.Image(type="pil", label="Segmentation Mask"),
        gr.Textbox(label="Bounding Box Coordinates")
    ],
    examples=[[example]],
    cache_examples=False,
    title="Microscopy Cell Segmentation with Stable Diffusion + LoCA",
    description="Upload a microscopy image and draw a bounding box to segment cell instances using an attention-guided diffusion model."
)

# 启动
if __name__ == "__main__":
    demo.queue().launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        show_error=True,
        ssr_mode=False
    )