app.py

# app.py  (HF Spaces: SDK=gradio)
import io, base64, numpy as np, torch, gradio as gr
from PIL import Image
from transformers import AutoImageProcessor, DepthProForDepthEstimation

device = "cuda" if torch.cuda.is_available() else "cpu"
_proc = None
_model = None

def _lazy_init():
    global _proc, _model
    if _proc is None:
        _proc = AutoImageProcessor.from_pretrained("apple/DepthPro-hf")
    if _model is None:
        _model = DepthProForDepthEstimation.from_pretrained("apple/DepthPro-hf").to(device).eval()

def _infer(pil_img: Image.Image):
    _lazy_init()
    H, W = pil_img.height, pil_img.width
    inputs = _proc(images=pil_img.convert("RGB"), return_tensors="pt").to(device)
    with torch.no_grad():
        outputs = _model(**inputs)
    post = _proc.post_process_depth_estimation(outputs, target_sizes=[(H, W)])[0]
    depth = post["predicted_depth"].float().cpu().numpy()
    fov   = float(post.get("field_of_view", 0.0))
    focal = float(post.get("focal_length", 0.0))
    return depth, H, W, fov, focal

# (A) API 함수: JSON 반환
def depth_api(img: Image.Image):
    depth, H, W, fov, focal = _infer(img)
    depth_b64 = base64.b64encode(depth.astype(np.float32).tobytes()).decode("ascii")
    return {
        "height": int(H),
        "width": int(W),
        "focal_px": float(focal),
        "field_of_view": float(fov),
        "depth_flat": depth_b64
    }

# (B) 프리뷰용 UI
def preview(img: Image.Image):
    depth, *_ = _infer(img)
    v = depth[np.isfinite(depth)]
    lo, hi = (np.percentile(v, 1), np.percentile(v, 99)) if v.size else (0, 1)
    norm = np.clip((depth - lo) / max(1e-6, hi - lo), 0, 1)
    return Image.fromarray((norm * 255).astype(np.uint8))

# 🔹 Blocks(UI) 만들기
with gr.Blocks() as ui:
    gr.Markdown("## DepthPro-hf (CPU, Free Space)\n- REST API: **POST /api/predict/depth** (JSON base64)")
    with gr.Row():
        inp = gr.Image(type="pil", label="Input")
        out = gr.Image(label="Depth (preview)")
    gr.Button("Run").click(preview, inp, out)

# 🔹 API 인터페이스 (REST 경로: /api/predict/depth)
api = gr.Interface(
    fn=depth_api,
    inputs=gr.Image(type="pil"),
    outputs=gr.JSON(),
    api_name="depth"
)

# ✅ 두 개를 하나의 앱으로 합치기
demo = gr.TabbedInterface([ui, api], tab_names=["UI", "api"])