andevs commited on
Commit
02c49fe
·
verified ·
1 Parent(s): 7cfb24c

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +371 -1
app.py CHANGED
@@ -1772,4 +1772,374 @@ async def process_bulk_images(
1772
  max_size=max_size, quality=95,
1773
  preserve_hair=preserve_hair,
1774
  use_alpha_matting=use_alpha_matting,
1775
- border_pixels=border_pixels
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1772
  max_size=max_size, quality=95,
1773
  preserve_hair=preserve_hair,
1774
  use_alpha_matting=use_alpha_matting,
1775
+ border_pixels=border_pixels, sharpen=sharpen,
1776
+ )
1777
+ scores.append(d['quality_score'])
1778
+ stem = Path(f.filename).stem or "image"
1779
+ out_stem = f"{stem}_remova"
1780
+ if out_stem in used_names:
1781
+ n = 2
1782
+ while f"{out_stem}_{n}" in used_names: n += 1
1783
+ out_stem = f"{out_stem}_{n}"
1784
+ used_names.add(out_stem)
1785
+ out_name = f"{out_stem}.{d['extension']}"
1786
+ zf.writestr(out_name, d['data'])
1787
+ count += 1
1788
+ except Exception as e:
1789
+ logger.error(f"Failed {f.filename}: {e}")
1790
+
1791
+ if count == 0: raise HTTPException(500, "No images processed")
1792
+
1793
+ zb.seek(0)
1794
+ pt = time.time() - t0
1795
+ avg = sum(scores) / len(scores) if scores else 0
1796
+ headers = {
1797
+ 'X-Processing-Time': f'{pt:.3f}',
1798
+ 'X-Model-Used': 'smart_routing',
1799
+ 'X-Image-Count': str(count),
1800
+ 'X-Avg-Quality-Score': f'{avg:.1f}',
1801
+ 'Content-Disposition': f'attachment; filename=processed_images_{int(time.time())}.zip',
1802
+ }
1803
+ return StreamingResponse(zb, media_type='application/zip', headers=headers)
1804
+ except HTTPException:
1805
+ raise
1806
+ except Exception as e:
1807
+ logger.error(f"Bulk error: {e}")
1808
+ raise HTTPException(500, str(e)[:200])
1809
+
1810
+
1811
+ @app.post("/api/process/frames-batch")
1812
+ async def process_frames_batch(
1813
+ files: List[UploadFile] = File(...),
1814
+ model: str = Form(DEFAULT_VIDEO_MODEL),
1815
+ transparent: bool = Form(True),
1816
+ session_id: Optional[str] = Query(None),
1817
+ smooth: bool = Query(True),
1818
+ ):
1819
+ if len(files) > VIDEO_MAX_BATCH:
1820
+ raise HTTPException(400, f"Max {VIDEO_MAX_BATCH} frames per batch")
1821
+
1822
+ mem = get_container_memory()
1823
+ if mem['rss_percent'] > VIDEO_RAM_REJECT_PERCENT:
1824
+ logger.warning(f"[frames-batch] RSS {mem['rss_percent']:.1f}% > {VIDEO_RAM_REJECT_PERCENT}%")
1825
+ return JSONResponse(
1826
+ status_code=503,
1827
+ content={
1828
+ 'error': 'Server memory high. Retry shortly.',
1829
+ 'rss_percent': round(mem['rss_percent'], 1),
1830
+ 'rss_mb': mem['process_rss_mb'],
1831
+ 'container_percent': round(mem['percent'], 1),
1832
+ 'retry_in': 10, 'cold': True,
1833
+ },
1834
+ headers={'Retry-After': '10', 'Cache-Control': 'no-store'},
1835
+ )
1836
+
1837
+ raw_frames = []
1838
+ for f in files:
1839
+ try:
1840
+ raw_frames.append(await f.read())
1841
+ except Exception as e:
1842
+ logger.warning(f"[frames-batch] read failed: {e}")
1843
+ raw_frames.append(b"")
1844
+
1845
+ loop = asyncio.get_event_loop()
1846
+
1847
+ def _worker():
1848
+ with _video_semaphore:
1849
+ return _process_frames_sync(
1850
+ raw_frames, VIDEO_FORCE_MODEL, transparent, session_id, smooth,
1851
+ )
1852
+
1853
+ try:
1854
+ results = await loop.run_in_executor(cpu_executor, _worker)
1855
+ except Exception as e:
1856
+ logger.error(f"[frames-batch] worker crashed: {e}")
1857
+ raise HTTPException(500, f"Frame processing failed: {str(e)[:200]}")
1858
+ finally:
1859
+ gc.collect()
1860
+ trim_memory()
1861
+
1862
+ used = next((r.get("model") for r in results if r.get("model")), VIDEO_FORCE_MODEL)
1863
+ smoothed_count = sum(1 for r in results if r.get("smoothed"))
1864
+ return {
1865
+ "frames": results, "count": len(results), "model": used,
1866
+ "transparent": transparent,
1867
+ "session_id": session_id if (smooth and session_id) else None,
1868
+ "smoothed_count": smoothed_count,
1869
+ "smooth_weight": ALPHA_SMOOTH_WEIGHT if smoothed_count else 0.0,
1870
+ }
1871
+
1872
+
1873
+ @app.post("/api/frames/session/reset")
1874
+ async def reset_frame_session(session_id: str = Query(...)):
1875
+ existed = _reset_session(session_id)
1876
+ return {"ok": True, "session_id": session_id, "existed": existed}
1877
+
1878
+
1879
+ @app.get("/api/frames/session/stats")
1880
+ async def frame_session_stats():
1881
+ return _alpha_cache_stats()
1882
+
1883
+
1884
+ @app.post("/api/editor/auto-crop")
1885
+ async def editor_auto_crop(file: UploadFile = File(...), padding: int = Query(0, ge=0, le=20)):
1886
+ try:
1887
+ contents = await file.read()
1888
+ img = Image.open(io.BytesIO(contents)).convert('RGBA')
1889
+ arr = np.array(img)
1890
+ alpha = arr[:, :, 3]
1891
+ ys, xs = np.where(alpha > 20)
1892
+ if len(ys) == 0:
1893
+ return JSONResponse({'success': False, 'error': 'No foreground found'})
1894
+ x1, x2 = int(xs.min()), int(xs.max()) + 1
1895
+ y1, y2 = int(ys.min()), int(ys.max()) + 1
1896
+ w, h = img.size
1897
+ if padding > 0:
1898
+ pw = int((x2 - x1) * padding / 100)
1899
+ ph = int((y2 - y1) * padding / 100)
1900
+ x1 = max(0, x1 - pw); y1 = max(0, y1 - ph)
1901
+ x2 = min(w, x2 + pw); y2 = min(h, y2 + ph)
1902
+ return {'success': True, 'x': x1, 'y': y1,
1903
+ 'width': x2 - x1, 'height': y2 - y1,
1904
+ 'original_width': w, 'original_height': h}
1905
+ except Exception as e:
1906
+ logger.error(f"auto-crop failed: {e}")
1907
+ return JSONResponse({'success': False, 'error': str(e)}, status_code=500)
1908
+
1909
+
1910
+ @app.post("/api/editor/inpaint")
1911
+ async def editor_inpaint(file: UploadFile = File(...), mask: UploadFile = File(...),
1912
+ method: str = Query('telea'), dilate: int = Query(3)):
1913
+ try:
1914
+ contents = await file.read()
1915
+ mask_contents = await mask.read()
1916
+ pil = Image.open(io.BytesIO(contents)).convert('RGBA')
1917
+ arr = np.array(pil)
1918
+ rgb = arr[:, :, :3].copy()
1919
+ alpha = arr[:, :, 3].copy()
1920
+ h, w = rgb.shape[:2]
1921
+ mask_pil = Image.open(io.BytesIO(mask_contents)).convert('L')
1922
+ if mask_pil.size != pil.size:
1923
+ mask_pil = mask_pil.resize(pil.size, Image.Resampling.LANCZOS)
1924
+ mask_arr = np.array(mask_pil)
1925
+ if dilate > 0:
1926
+ mask_arr = cv2.dilate(mask_arr, np.ones((dilate, dilate), np.uint8), iterations=1)
1927
+ if method == 'ns':
1928
+ inpainted = cv2.inpaint(rgb, mask_arr, 3, cv2.INPAINT_NS)
1929
+ else:
1930
+ inpainted = cv2.inpaint(rgb, mask_arr, 3, cv2.INPAINT_TELEA)
1931
+ mask_float = mask_arr.astype(np.float32) / 255.0
1932
+ from scipy.ndimage import gaussian_filter
1933
+ mask_blur = gaussian_filter(mask_float, sigma=1.5)
1934
+ mask_3ch = mask_blur[:, :, None]
1935
+ blended = rgb * (1 - mask_3ch) + inpainted * mask_3ch
1936
+ blended = np.clip(blended, 0, 255).astype(np.uint8)
1937
+ result = np.zeros((h, w, 4), dtype=np.uint8)
1938
+ result[:, :, :3] = blended
1939
+ result[:, :, 3] = alpha
1940
+ out = Image.fromarray(result, 'RGBA')
1941
+ buf = io.BytesIO()
1942
+ out.save(buf, format='PNG', optimize=True)
1943
+ buf.seek(0)
1944
+ return StreamingResponse(buf, media_type='image/png')
1945
+ except Exception as e:
1946
+ logger.error(f"inpaint failed: {e}")
1947
+ raise HTTPException(500, f"Inpainting failed: {str(e)[:200]}")
1948
+
1949
+
1950
+ @app.get("/api/warmup")
1951
+ async def warmup():
1952
+ return {"status": "warm", "ts": datetime.now().isoformat()}
1953
+
1954
+
1955
+ @app.get("/api/debug/memory")
1956
+ async def debug_memory():
1957
+ mem = get_container_memory()
1958
+ stats = model_manager.get_stats()
1959
+ return {
1960
+ 'container': mem,
1961
+ 'process_rss_mb': mem['process_rss_mb'],
1962
+ 'rss_percent': round(mem['rss_percent'], 1),
1963
+ 'cgroup_percent': round(mem['percent'], 1),
1964
+ 'note': 'rss_percent is the real number. cgroup_percent includes page cache.',
1965
+ 'models_loaded': stats['loaded_models'],
1966
+ 'models_count': stats['model_count'],
1967
+ 'models_warmed': stats['warmed'],
1968
+ 'use_counts': stats['use_counts'],
1969
+ 'protected': stats['protected'],
1970
+ 'alpha_cache': _alpha_cache_stats(),
1971
+ 'video_reject_pct': VIDEO_RAM_REJECT_PERCENT,
1972
+ 'torch_loaded': 'torch' in sys.modules,
1973
+ }
1974
+
1975
+
1976
+ @app.get("/api/debug/tmp")
1977
+ async def debug_tmp():
1978
+ items = []
1979
+ try:
1980
+ for root, dirs, files in os.walk('/tmp'):
1981
+ for f in files:
1982
+ p = os.path.join(root, f)
1983
+ try:
1984
+ items.append({
1985
+ 'path': p,
1986
+ 'size_mb': round(os.path.getsize(p) / 1048576, 2),
1987
+ 'mtime': os.path.getmtime(p),
1988
+ })
1989
+ except Exception:
1990
+ pass
1991
+ except Exception:
1992
+ pass
1993
+ return {
1994
+ 'count': len(items),
1995
+ 'total_mb': round(sum(i['size_mb'] for i in items), 2),
1996
+ 'items': sorted(items, key=lambda x: -x['size_mb'])[:50],
1997
+ }
1998
+
1999
+
2000
+ @app.get("/api/debug/env")
2001
+ async def debug_env():
2002
+ mem = get_container_memory()
2003
+ return {
2004
+ 'total_mb': mem['total_mb'],
2005
+ 'used_mb': mem['used_mb'],
2006
+ 'available_mb': mem['available_mb'],
2007
+ 'container_percent': round(mem['percent'], 1),
2008
+ 'process_rss_mb': mem['process_rss_mb'],
2009
+ 'rss_percent': round(mem['rss_percent'], 1),
2010
+ 'memory_source': mem['source'],
2011
+ 'models_loaded': list(model_manager.models.keys()),
2012
+ 'models_warmed': dict(model_manager.warmed),
2013
+ 'active_models': sorted(VALID_REMBG_MODELS),
2014
+ 'dormant_models': sorted(DORMANT_MODELS.keys()),
2015
+ 'remove_func_ready': remove_func is not None,
2016
+ 'ai_available': AI_AVAILABLE,
2017
+ 'default_model': default_model_name,
2018
+ 'video_force_model': VIDEO_FORCE_MODEL,
2019
+ 'video_ram_reject_percent': VIDEO_RAM_REJECT_PERCENT,
2020
+ 'video_max_batch': VIDEO_MAX_BATCH,
2021
+ 'preload_models': [n for n, _ in PRELOAD_MODELS],
2022
+ 'torch_loaded': 'torch' in sys.modules,
2023
+ 'onnx_threads': 1,
2024
+ 'router': 'v32.6.3 — app(3).py router, 4 active + 6 dormant, RAM optimized',
2025
+ }
2026
+
2027
+
2028
+ @app.get("/")
2029
+ async def home():
2030
+ stats = model_manager.get_stats()
2031
+ mem = get_container_memory()
2032
+ warmed = sum(1 for w in stats['warmed'].values() if w)
2033
+ return HTMLResponse(f"""
2034
+ <!DOCTYPE html><html><head><meta charset="UTF-8">
2035
+ <title>Background Remover v32.6.3</title>
2036
+ <style>body{{font-family:system-ui;background:#0A0A0A;color:#fff;
2037
+ min-height:100vh;display:flex;align-items:center;justify-content:center;
2038
+ margin:0;padding:20px}}.c{{max-width:900px;background:#111;
2039
+ border-radius:20px;padding:40px;text-align:center;
2040
+ border:1px solid #222}}.s{{padding:15px;border-radius:12px;
2041
+ margin:20px 0;background:#1a1a1a}}a{{color:#7ED957}}</style>
2042
+ </head><body><div class="c">
2043
+ <h1>Background Remover v32.6.3</h1>
2044
+ <div class="s">AI online. {len(stats['loaded_models'])} models ({warmed} warmed).
2045
+ RSS {mem['rss_percent']:.0f}% ({mem['process_rss_mb']}/{mem['total_mb']}MB)</div>
2046
+ <p>Active: silueta · u2netp · isnet · birefnet-lite</p>
2047
+ <p>Dormant: u2net · human_seg · anime · birefnet · cloth_seg · sam</p>
2048
+ <p>Router: flat→flood_fill · human→isnet · small→u2netp · complex→birefnet</p>
2049
+ <p><a href="/docs">API Docs</a> ·
2050
+ <a href="/api/models">Models</a> ·
2051
+ <a href="/api/debug/memory">Memory</a> ·
2052
+ <a href="/api/debug/tmp">Disk</a> ·
2053
+ <a href="/api/debug/env">Env</a></p>
2054
+ </div></body></html>
2055
+ """)
2056
+
2057
+
2058
+ @app.get("/api/health")
2059
+ async def health_check():
2060
+ stats = model_manager.get_stats()
2061
+ mem = get_container_memory()
2062
+
2063
+ preload_names = [n for n, _ in PRELOAD_MODELS]
2064
+ all_loaded = all(n in stats['loaded_models'] for n in preload_names)
2065
+ all_warmed = all(stats['warmed'].get(n, False) for n in preload_names)
2066
+ ram_ok = mem['rss_percent'] < (VIDEO_RAM_REJECT_PERCENT - 5)
2067
+
2068
+ if not AI_AVAILABLE:
2069
+ status = 'error'
2070
+ elif not all_loaded:
2071
+ status = 'degraded'
2072
+ elif not all_warmed:
2073
+ status = 'warming'
2074
+ elif not ram_ok:
2075
+ status = 'degraded'
2076
+ else:
2077
+ status = 'healthy'
2078
+
2079
+ return {
2080
+ 'status': status,
2081
+ 'ai_available': AI_AVAILABLE,
2082
+ 'models_loaded': stats['loaded_models'],
2083
+ 'models_count': stats['model_count'],
2084
+ 'models_warmed': stats['warmed'],
2085
+ 'all_loaded': all_loaded,
2086
+ 'all_warmed': all_warmed,
2087
+ 'active_models': sorted(VALID_REMBG_MODELS),
2088
+ 'dormant_models': sorted(DORMANT_MODELS.keys()),
2089
+ 'protected_models': stats['protected'],
2090
+ 'rss_percent': round(mem['rss_percent'], 1),
2091
+ 'rss_mb': mem['process_rss_mb'],
2092
+ 'container_percent': round(mem['percent'], 1),
2093
+ 'memory_total_mb': mem['total_mb'],
2094
+ 'memory_source': mem['source'],
2095
+ 'video_ram_reject_percent': VIDEO_RAM_REJECT_PERCENT,
2096
+ 'video_max_batch': VIDEO_MAX_BATCH,
2097
+ 'video_force_model': VIDEO_FORCE_MODEL,
2098
+ 'version': '32.6.3',
2099
+ 'preload_models': preload_names,
2100
+ 'alpha_cache_ttl': ALPHA_CACHE_TTL_SECONDS,
2101
+ 'alpha_purge_interval': ALPHA_PURGE_INTERVAL,
2102
+ 'alpha_smoothing': _alpha_cache_stats(),
2103
+ 'router_version': 'v32.6.3',
2104
+ 'onnx_threads': 1,
2105
+ 'torch_loaded': 'torch' in sys.modules,
2106
+ }
2107
+
2108
+
2109
+ @app.exception_handler(HTTPException)
2110
+ async def http_exc(request, exc):
2111
+ return JSONResponse(status_code=exc.status_code,
2112
+ content={'error': exc.detail, 'status_code': exc.status_code})
2113
+
2114
+
2115
+ @app.exception_handler(Exception)
2116
+ async def gen_exc(request, exc):
2117
+ logger.error(f"Unhandled: {exc}")
2118
+ return JSONResponse(status_code=500,
2119
+ content={'error': 'Internal error', 'detail': str(exc)[:200]})
2120
+
2121
+
2122
+ @app.on_event("startup")
2123
+ async def startup():
2124
+ logger.info("=" * 60)
2125
+ logger.info("v32.6.3 ready")
2126
+ logger.info(f"Active: {sorted(VALID_REMBG_MODELS)}")
2127
+ logger.info(f"Dormant: {sorted(DORMANT_MODELS.keys())}")
2128
+ logger.info("Dormant requests reroute through complexity router")
2129
+ logger.info("ONNX threads: 1 (matches Dockerfile)")
2130
+ logger.info("Warm res: silueta 384, u2netp 384, isnet 128, birefnet 128")
2131
+ logger.info("Birefnet cap: 512, isnet cap: 576")
2132
+ logger.info("=" * 60)
2133
+ _log_rss("startup")
2134
+
2135
+
2136
+ @app.on_event("shutdown")
2137
+ async def shutdown():
2138
+ cpu_executor.shutdown(wait=True)
2139
+
2140
+
2141
+ if __name__ == '__main__':
2142
+ import uvicorn
2143
+ uvicorn.run('app:app', host='0.0.0.0',
2144
+ port=int(os.environ.get('PORT', 7860)),
2145
+ timeout_keep_alive=300, log_level='info', reload=False)