Spaces:
Sleeping
Sleeping
Delete main.py
Browse files
main.py
DELETED
|
@@ -1,671 +0,0 @@
|
|
| 1 |
-
import os
|
| 2 |
-
import shutil
|
| 3 |
-
import asyncio
|
| 4 |
-
import subprocess
|
| 5 |
-
import numpy as np
|
| 6 |
-
from pathlib import Path
|
| 7 |
-
from typing import Optional
|
| 8 |
-
from contextlib import asynccontextmanager
|
| 9 |
-
|
| 10 |
-
from fastapi import FastAPI, UploadFile, File, HTTPException, BackgroundTasks
|
| 11 |
-
from fastapi.middleware.cors import CORSMiddleware
|
| 12 |
-
from fastapi.responses import FileResponse
|
| 13 |
-
|
| 14 |
-
# ── paths ──────────────────────────────────────────────────────────────────────
|
| 15 |
-
TMP = Path("/tmp/clipcut")
|
| 16 |
-
TMP.mkdir(parents=True, exist_ok=True)
|
| 17 |
-
os.environ["INSIGHTFACE_HOME"] = "/tmp"
|
| 18 |
-
|
| 19 |
-
# ── lazy model ─────────────────────────────────────────────────────────────────
|
| 20 |
-
_face_app = None
|
| 21 |
-
|
| 22 |
-
def get_face_app():
|
| 23 |
-
global _face_app
|
| 24 |
-
if _face_app is None:
|
| 25 |
-
from insightface.app import FaceAnalysis
|
| 26 |
-
_face_app = FaceAnalysis(name="buffalo_sc", root="/tmp/insightface",
|
| 27 |
-
providers=["CPUExecutionProvider"])
|
| 28 |
-
_face_app.prepare(ctx_id=-1, det_size=(640, 640))
|
| 29 |
-
return _face_app
|
| 30 |
-
|
| 31 |
-
# ── job store ──────────────────────────────────────────────────────────────────
|
| 32 |
-
jobs: dict[str, dict] = {}
|
| 33 |
-
|
| 34 |
-
@asynccontextmanager
|
| 35 |
-
async def lifespan(app):
|
| 36 |
-
yield
|
| 37 |
-
|
| 38 |
-
app = FastAPI(lifespan=lifespan)
|
| 39 |
-
app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
|
| 40 |
-
|
| 41 |
-
# ── utils ──────────────────────────────────────────────────────────────────────
|
| 42 |
-
|
| 43 |
-
def cosine_sim(a, b) -> float:
|
| 44 |
-
a, b = a.flatten(), b.flatten()
|
| 45 |
-
d = np.linalg.norm(a) * np.linalg.norm(b)
|
| 46 |
-
return float(np.dot(a, b) / d) if d > 0 else 0.0
|
| 47 |
-
|
| 48 |
-
def clothing_hist(frame, bbox) -> Optional[np.ndarray]:
|
| 49 |
-
import cv2
|
| 50 |
-
x1, y1, x2, y2 = int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])
|
| 51 |
-
region = frame[y1 + int((y2-y1)*0.3):y2, x1:x2]
|
| 52 |
-
if region.size == 0:
|
| 53 |
-
return None
|
| 54 |
-
hsv = cv2.cvtColor(region, cv2.COLOR_BGR2HSV)
|
| 55 |
-
h = cv2.calcHist([hsv], [0,1], None, [16,16], [0,180,0,256])
|
| 56 |
-
cv2.normalize(h, h)
|
| 57 |
-
return h.flatten()
|
| 58 |
-
|
| 59 |
-
def hist_sim(h1, h2) -> float:
|
| 60 |
-
import cv2
|
| 61 |
-
if h1 is None or h2 is None:
|
| 62 |
-
return 0.0
|
| 63 |
-
return float(cv2.compareHist(h1.astype(np.float32).reshape(-1,1),
|
| 64 |
-
h2.astype(np.float32).reshape(-1,1),
|
| 65 |
-
cv2.HISTCMP_CORREL))
|
| 66 |
-
|
| 67 |
-
def extract_ref_embedding(ref_path: Path) -> Optional[np.ndarray]:
|
| 68 |
-
import cv2
|
| 69 |
-
img = cv2.imread(str(ref_path))
|
| 70 |
-
if img is None:
|
| 71 |
-
return None
|
| 72 |
-
faces = get_face_app().get(img)
|
| 73 |
-
if not faces:
|
| 74 |
-
return None
|
| 75 |
-
largest = max(faces, key=lambda f: (f.bbox[2]-f.bbox[0])*(f.bbox[3]-f.bbox[1]))
|
| 76 |
-
return largest.normed_embedding
|
| 77 |
-
|
| 78 |
-
def video_info(path: Path) -> dict:
|
| 79 |
-
import json
|
| 80 |
-
r = subprocess.run([
|
| 81 |
-
"ffprobe", "-v", "error", "-select_streams", "v:0",
|
| 82 |
-
"-show_entries", "stream=width,height,r_frame_rate",
|
| 83 |
-
"-show_entries", "format=duration", "-of", "json", str(path)
|
| 84 |
-
], capture_output=True, text=True)
|
| 85 |
-
d = json.loads(r.stdout)
|
| 86 |
-
s = d.get("streams", [{}])[0]
|
| 87 |
-
num, den = s.get("r_frame_rate", "30/1").split("/")
|
| 88 |
-
return {
|
| 89 |
-
"fps": float(num)/float(den),
|
| 90 |
-
"duration": float(d.get("format",{}).get("duration", 0)),
|
| 91 |
-
"width": int(s.get("width", 1080)),
|
| 92 |
-
"height": int(s.get("height", 1920))
|
| 93 |
-
}
|
| 94 |
-
|
| 95 |
-
def crop_size(w, h) -> tuple[int, int]:
|
| 96 |
-
if (w/h) > (9/16):
|
| 97 |
-
ch = h - h%2
|
| 98 |
-
cw = int(ch*9/16) - int(ch*9/16)%2
|
| 99 |
-
else:
|
| 100 |
-
cw = w - w%2
|
| 101 |
-
ch = int(cw*16/9) - int(cw*16/9)%2
|
| 102 |
-
return cw, ch
|
| 103 |
-
|
| 104 |
-
def clamp(cx, cy, cw, ch, fw, fh) -> tuple[int, int]:
|
| 105 |
-
x = max(0, min(cx - cw//2, fw - cw))
|
| 106 |
-
y = max(0, min(cy - ch//2, fh - ch))
|
| 107 |
-
return x, y
|
| 108 |
-
|
| 109 |
-
# ── scan video ─────────────────────────────────────────────────────────────────
|
| 110 |
-
|
| 111 |
-
def scan_video(video_path: Path, ref_emb: np.ndarray, fps: float,
|
| 112 |
-
job_id: str, p0=15, p1=82, label="Scanning"):
|
| 113 |
-
import cv2
|
| 114 |
-
import time
|
| 115 |
-
fa = get_face_app()
|
| 116 |
-
FACE_THRESH = 0.20
|
| 117 |
-
BODY_THRESH = 0.55
|
| 118 |
-
BODY_WINDOW = fps * 4
|
| 119 |
-
|
| 120 |
-
cap = cv2.VideoCapture(str(video_path))
|
| 121 |
-
total = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
| 122 |
-
fw = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
| 123 |
-
fh = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
| 124 |
-
|
| 125 |
-
timestamps, positions = [], {}
|
| 126 |
-
ref_hist = None
|
| 127 |
-
last_face_frame = -9999
|
| 128 |
-
idx = 0
|
| 129 |
-
start_time = time.time()
|
| 130 |
-
preview_dir = TMP / job_id
|
| 131 |
-
PREVIEW_EVERY = max(1, total // 20) # save ~20 previews during scan
|
| 132 |
-
|
| 133 |
-
while True:
|
| 134 |
-
# ── Check cancel flag ──
|
| 135 |
-
if jobs.get(job_id, {}).get("cancelled"):
|
| 136 |
-
cap.release()
|
| 137 |
-
return [], {}
|
| 138 |
-
|
| 139 |
-
ret, frame = cap.read()
|
| 140 |
-
if not ret:
|
| 141 |
-
break
|
| 142 |
-
|
| 143 |
-
if idx % 2 == 0:
|
| 144 |
-
faces = fa.get(frame)
|
| 145 |
-
matched = False
|
| 146 |
-
cx_r, cy_r = 0.5, 0.42
|
| 147 |
-
|
| 148 |
-
for face in faces:
|
| 149 |
-
if cosine_sim(face.normed_embedding, ref_emb) >= FACE_THRESH:
|
| 150 |
-
matched = True
|
| 151 |
-
last_face_frame = idx
|
| 152 |
-
x1, y1, x2, y2 = face.bbox
|
| 153 |
-
fh_box = y2 - y1
|
| 154 |
-
cx_r = ((x1+x2)/2) / fw
|
| 155 |
-
cy_r = min(0.88, ((y1+y2)/2 + fh_box*1.2) / fh)
|
| 156 |
-
h = clothing_hist(frame, face.bbox)
|
| 157 |
-
if h is not None:
|
| 158 |
-
ref_hist = h
|
| 159 |
-
break
|
| 160 |
-
|
| 161 |
-
if not matched and ref_hist is not None and (idx - last_face_frame) <= BODY_WINDOW:
|
| 162 |
-
best_s, best_bbox = 0.0, None
|
| 163 |
-
for face in faces:
|
| 164 |
-
s = hist_sim(ref_hist, clothing_hist(frame, face.bbox))
|
| 165 |
-
if s > best_s:
|
| 166 |
-
best_s, best_bbox = s, face.bbox
|
| 167 |
-
if best_s >= BODY_THRESH and best_bbox is not None:
|
| 168 |
-
matched = True
|
| 169 |
-
x1, y1, x2, y2 = best_bbox
|
| 170 |
-
cx_r = ((x1+x2)/2) / fw
|
| 171 |
-
cy_r = min(0.88, ((y1+y2)/2) / fh)
|
| 172 |
-
|
| 173 |
-
if matched:
|
| 174 |
-
ts = idx / fps
|
| 175 |
-
timestamps.append(ts)
|
| 176 |
-
positions[ts] = (cx_r, cy_r)
|
| 177 |
-
|
| 178 |
-
# ── Save preview frame every N frames ──
|
| 179 |
-
if idx % PREVIEW_EVERY == 0 and frame is not None:
|
| 180 |
-
try:
|
| 181 |
-
preview_path = preview_dir / "preview.jpg"
|
| 182 |
-
small = cv2.resize(frame, (360, int(fh * 360 / fw)))
|
| 183 |
-
cv2.imwrite(str(preview_path), small, [cv2.IMWRITE_JPEG_QUALITY, 70])
|
| 184 |
-
except Exception:
|
| 185 |
-
pass
|
| 186 |
-
|
| 187 |
-
# ── ETA calculation ──
|
| 188 |
-
elapsed = time.time() - start_time
|
| 189 |
-
prange = p1 - p0
|
| 190 |
-
prog = p0 + int((idx / max(total, 1)) * prange)
|
| 191 |
-
if idx > 0 and elapsed > 0:
|
| 192 |
-
fps_proc = idx / elapsed
|
| 193 |
-
frames_left = total - idx
|
| 194 |
-
eta_sec = int(frames_left / fps_proc)
|
| 195 |
-
if eta_sec >= 60:
|
| 196 |
-
eta_str = f"{eta_sec//60}m {eta_sec%60}s"
|
| 197 |
-
else:
|
| 198 |
-
eta_str = f"{eta_sec}s"
|
| 199 |
-
else:
|
| 200 |
-
eta_str = "calculating..."
|
| 201 |
-
|
| 202 |
-
jobs[job_id]["progress"] = prog
|
| 203 |
-
jobs[job_id]["status_text"] = f"{label}: {idx}/{total} frames"
|
| 204 |
-
jobs[job_id]["eta"] = eta_str
|
| 205 |
-
idx += 1
|
| 206 |
-
|
| 207 |
-
cap.release()
|
| 208 |
-
return timestamps, positions
|
| 209 |
-
|
| 210 |
-
# ── intervals ──────────────────────────────────────────────────────────────────
|
| 211 |
-
|
| 212 |
-
def to_intervals(timestamps, fps, gap=3.0, min_dur=0.3):
|
| 213 |
-
if not timestamps:
|
| 214 |
-
return []
|
| 215 |
-
fw = 1.0/fps
|
| 216 |
-
intervals, start, end = [], timestamps[0], timestamps[0]+fw
|
| 217 |
-
for ts in timestamps[1:]:
|
| 218 |
-
if ts - end <= gap:
|
| 219 |
-
end = ts + fw
|
| 220 |
-
else:
|
| 221 |
-
if end-start >= min_dur:
|
| 222 |
-
intervals.append((start, end))
|
| 223 |
-
start, end = ts, ts+fw
|
| 224 |
-
if end-start >= min_dur:
|
| 225 |
-
intervals.append((start, end))
|
| 226 |
-
return intervals
|
| 227 |
-
|
| 228 |
-
# ── cut clips with dynamic crop ────────────────────────────────────────────────
|
| 229 |
-
|
| 230 |
-
def cut_clip(video_path, start, end, positions, vi, out_path) -> bool:
|
| 231 |
-
w, h = vi["width"], vi["height"]
|
| 232 |
-
fps = vi["fps"]
|
| 233 |
-
cw, ch = crop_size(w, h)
|
| 234 |
-
dur = end - start
|
| 235 |
-
|
| 236 |
-
# Gather positions for this interval
|
| 237 |
-
pts = {ts: pos for ts, pos in positions.items() if start-0.1 <= ts <= end+0.1}
|
| 238 |
-
|
| 239 |
-
if len(pts) >= 2:
|
| 240 |
-
sorted_ts = sorted(pts.keys())
|
| 241 |
-
rel = [(ts-start, pts[ts]) for ts in sorted_ts]
|
| 242 |
-
|
| 243 |
-
def interp(t):
|
| 244 |
-
if t <= rel[0][0]: return rel[0][1]
|
| 245 |
-
if t >= rel[-1][0]: return rel[-1][1]
|
| 246 |
-
for i in range(len(rel)-1):
|
| 247 |
-
t0, p0 = rel[i]; t1, p1 = rel[i+1]
|
| 248 |
-
if t0 <= t <= t1:
|
| 249 |
-
a = (t-t0)/max(t1-t0, 1e-6)
|
| 250 |
-
return (p0[0]+a*(p1[0]-p0[0]), p0[1]+a*(p1[1]-p0[1]))
|
| 251 |
-
return rel[-1][1]
|
| 252 |
-
|
| 253 |
-
# Build per-frame positions
|
| 254 |
-
step = 1.0/fps
|
| 255 |
-
raw = []
|
| 256 |
-
t = 0.0
|
| 257 |
-
while t <= dur+step:
|
| 258 |
-
cx_r, cy_r = interp(t)
|
| 259 |
-
x, y = clamp(int(cx_r*w), int(cy_r*h), cw, ch, w, h)
|
| 260 |
-
raw.append((t, x, y))
|
| 261 |
-
t += step
|
| 262 |
-
|
| 263 |
-
# Smooth with small window + weight recent frames more to reduce lag
|
| 264 |
-
win = max(1, int(fps * 0.12)) # 0.12s — fast response to movement
|
| 265 |
-
smoothed = []
|
| 266 |
-
for i, (t, x, y) in enumerate(raw):
|
| 267 |
-
i0, i1 = max(0, i-win), min(len(raw), i+win+1)
|
| 268 |
-
weights = [1.0 / (abs(j-i)+1) for j in range(i0, i1)]
|
| 269 |
-
xs = [raw[j][1] for j in range(i0, i1)]
|
| 270 |
-
ys = [raw[j][2] for j in range(i0, i1)]
|
| 271 |
-
sw = sum(weights)
|
| 272 |
-
sx = int(sum(wt*xv for wt,xv in zip(weights,xs)) / sw)
|
| 273 |
-
sy = int(sum(wt*yv for wt,yv in zip(weights,ys)) / sw)
|
| 274 |
-
smoothed.append((t, sx, sy))
|
| 275 |
-
|
| 276 |
-
# Write sendcmd file
|
| 277 |
-
sc = out_path.parent / f"sc_{out_path.stem}.txt"
|
| 278 |
-
with open(sc, "w") as f:
|
| 279 |
-
for i, (t, x, y) in enumerate(smoothed):
|
| 280 |
-
f.write(f"{t:.4f} crop x {x}; crop y {y};\n")
|
| 281 |
-
|
| 282 |
-
cmd = [
|
| 283 |
-
"ffmpeg", "-y", "-ss", str(start), "-i", str(video_path),
|
| 284 |
-
"-t", str(dur),
|
| 285 |
-
"-vf", f"crop={cw}:{ch}:0:0,sendcmd=f={sc},scale=1080:1920",
|
| 286 |
-
"-c:v", "libx264", "-preset", "fast", "-crf", "23",
|
| 287 |
-
"-c:a", "aac", "-b:a", "128k", "-movflags", "+faststart",
|
| 288 |
-
str(out_path)
|
| 289 |
-
]
|
| 290 |
-
r = subprocess.run(cmd, capture_output=True)
|
| 291 |
-
if sc.exists(): sc.unlink()
|
| 292 |
-
if r.returncode == 0 and out_path.exists():
|
| 293 |
-
return True
|
| 294 |
-
|
| 295 |
-
# Static fallback
|
| 296 |
-
vals = list(pts.values()) if pts else [(0.5, 0.42)]
|
| 297 |
-
cx_r = float(np.median([v[0] for v in vals]))
|
| 298 |
-
cy_r = float(np.median([v[1] for v in vals]))
|
| 299 |
-
x, y = clamp(int(cx_r*w), int(cy_r*h), cw, ch, w, h)
|
| 300 |
-
cmd = [
|
| 301 |
-
"ffmpeg", "-y", "-ss", str(start), "-i", str(video_path),
|
| 302 |
-
"-t", str(dur),
|
| 303 |
-
"-vf", f"crop={cw}:{ch}:{x}:{y},scale=1080:1920",
|
| 304 |
-
"-c:v", "libx264", "-preset", "fast", "-crf", "23",
|
| 305 |
-
"-c:a", "aac", "-b:a", "128k", "-movflags", "+faststart",
|
| 306 |
-
str(out_path)
|
| 307 |
-
]
|
| 308 |
-
r = subprocess.run(cmd, capture_output=True)
|
| 309 |
-
return r.returncode == 0 and out_path.exists()
|
| 310 |
-
|
| 311 |
-
def merge_clips(clips, out_path) -> bool:
|
| 312 |
-
if not clips: return False
|
| 313 |
-
if len(clips) == 1:
|
| 314 |
-
shutil.copy(clips[0], out_path)
|
| 315 |
-
return True
|
| 316 |
-
lst = out_path.parent / "concat.txt"
|
| 317 |
-
lst.write_text("\n".join(f"file '{c}'" for c in clips))
|
| 318 |
-
r = subprocess.run(["ffmpeg", "-y", "-f", "concat", "-safe", "0",
|
| 319 |
-
"-i", str(lst), "-c", "copy", str(out_path)],
|
| 320 |
-
capture_output=True)
|
| 321 |
-
return r.returncode == 0
|
| 322 |
-
|
| 323 |
-
# ── pipeline ───────────────────────────────────────────────────────────────────
|
| 324 |
-
|
| 325 |
-
async def process_video(job_id: str, video_path: Path, ref_path: Path):
|
| 326 |
-
job = jobs[job_id]
|
| 327 |
-
try:
|
| 328 |
-
job["status"] = "processing"
|
| 329 |
-
job["status_text"] = "Analyzing reference photo..."
|
| 330 |
-
job["progress"] = 5
|
| 331 |
-
|
| 332 |
-
ref_emb = extract_ref_embedding(ref_path)
|
| 333 |
-
if ref_emb is None:
|
| 334 |
-
job["status"] = "failed"
|
| 335 |
-
job["error"] = "No face detected in reference photo. Use a clear front-facing photo."
|
| 336 |
-
return
|
| 337 |
-
|
| 338 |
-
job["status_text"] = "Reading video..."
|
| 339 |
-
job["progress"] = 10
|
| 340 |
-
vi = video_info(video_path)
|
| 341 |
-
fps, duration = vi["fps"], vi["duration"]
|
| 342 |
-
|
| 343 |
-
loop = asyncio.get_event_loop()
|
| 344 |
-
timestamps, positions = await loop.run_in_executor(
|
| 345 |
-
None, lambda: scan_video(video_path, ref_emb, fps, job_id))
|
| 346 |
-
|
| 347 |
-
if not timestamps or jobs[job_id].get("cancelled"):
|
| 348 |
-
if jobs[job_id].get("cancelled"):
|
| 349 |
-
job["status"] = "cancelled"
|
| 350 |
-
job["status_text"] = "Cancelled."
|
| 351 |
-
return
|
| 352 |
-
job["status"] = "failed"
|
| 353 |
-
job["error"] = "Could not find the reference person in the video. Try a clearer photo."
|
| 354 |
-
return
|
| 355 |
-
|
| 356 |
-
intervals = to_intervals(timestamps, fps)
|
| 357 |
-
if not intervals:
|
| 358 |
-
job["status"] = "failed"
|
| 359 |
-
job["error"] = "Person found but clips too short to extract."
|
| 360 |
-
return
|
| 361 |
-
|
| 362 |
-
out_dir = video_path.parent
|
| 363 |
-
clips = []
|
| 364 |
-
total_intervals = len(intervals)
|
| 365 |
-
|
| 366 |
-
for i, (s, e) in enumerate(intervals):
|
| 367 |
-
job["status_text"] = f"Cutting clip {i+1} of {total_intervals}..."
|
| 368 |
-
job["progress"] = 83 + int((i / max(total_intervals, 1)) * 10)
|
| 369 |
-
op = out_dir / f"clip_{i:04d}.mp4"
|
| 370 |
-
if cut_clip(video_path, s, e, positions, vi, op):
|
| 371 |
-
clips.append(op)
|
| 372 |
-
|
| 373 |
-
if not clips:
|
| 374 |
-
job["status"] = "failed"
|
| 375 |
-
job["error"] = "Failed to cut clips."
|
| 376 |
-
return
|
| 377 |
-
|
| 378 |
-
job["status_text"] = f"Merging {len(clips)} clip(s) into final video..."
|
| 379 |
-
job["progress"] = 94
|
| 380 |
-
|
| 381 |
-
final = out_dir / "result.mp4"
|
| 382 |
-
if not merge_clips(clips, final):
|
| 383 |
-
job["status"] = "failed"
|
| 384 |
-
job["error"] = "Failed to merge clips."
|
| 385 |
-
return
|
| 386 |
-
|
| 387 |
-
job["status_text"] = "Finalizing video..."
|
| 388 |
-
job["progress"] = 98
|
| 389 |
-
|
| 390 |
-
total_kept = sum(e-s for s,e in intervals)
|
| 391 |
-
job.update({
|
| 392 |
-
"status": "done", "progress": 100, "status_text": "Done!",
|
| 393 |
-
"result_path": str(final),
|
| 394 |
-
"stats": {"duration": round(duration,1),
|
| 395 |
-
"extracted": round(total_kept,1),
|
| 396 |
-
"clips": len(intervals)}
|
| 397 |
-
})
|
| 398 |
-
|
| 399 |
-
except Exception as e:
|
| 400 |
-
import traceback
|
| 401 |
-
job["status"] = "failed"
|
| 402 |
-
job["error"] = f"Error: {str(e)}"
|
| 403 |
-
print(traceback.format_exc())
|
| 404 |
-
|
| 405 |
-
# ── routes ─────────────────────────────────────────────────────────────────────
|
| 406 |
-
|
| 407 |
-
@app.get("/health")
|
| 408 |
-
def health():
|
| 409 |
-
return {"status": "ok", "face_recognition": True, "device": "cpu"}
|
| 410 |
-
|
| 411 |
-
@app.post("/upload-chunk")
|
| 412 |
-
async def upload_chunk(job_id: str, chunk_index: int, total_chunks: int,
|
| 413 |
-
file_type: str, chunk: UploadFile = File(...)):
|
| 414 |
-
job_dir = TMP / job_id
|
| 415 |
-
job_dir.mkdir(parents=True, exist_ok=True)
|
| 416 |
-
chunk_dir = job_dir / f"{file_type}_chunks"
|
| 417 |
-
chunk_dir.mkdir(exist_ok=True)
|
| 418 |
-
(chunk_dir / f"{chunk_index:06d}").write_bytes(await chunk.read())
|
| 419 |
-
|
| 420 |
-
received = len(list(chunk_dir.glob("*")))
|
| 421 |
-
if received == total_chunks:
|
| 422 |
-
ext = ".mp4" if file_type == "video" else ".jpg"
|
| 423 |
-
final = job_dir / f"{file_type}{ext}"
|
| 424 |
-
with open(final, "wb") as out:
|
| 425 |
-
for i in range(total_chunks):
|
| 426 |
-
out.write((chunk_dir / f"{i:06d}").read_bytes())
|
| 427 |
-
shutil.rmtree(chunk_dir)
|
| 428 |
-
return {"status": "complete"}
|
| 429 |
-
return {"status": "partial", "received": received, "total": total_chunks}
|
| 430 |
-
|
| 431 |
-
@app.post("/process")
|
| 432 |
-
async def process(job_id: str, background_tasks: BackgroundTasks):
|
| 433 |
-
job_dir = TMP / job_id
|
| 434 |
-
video_path = next(
|
| 435 |
-
(job_dir / f"video{ext}" for ext in [".mp4",".mov",".avi",".mkv",".webm"]
|
| 436 |
-
if (job_dir / f"video{ext}").exists()), None)
|
| 437 |
-
ref_path = job_dir / "reference.jpg"
|
| 438 |
-
if not video_path:
|
| 439 |
-
raise HTTPException(400, "Video not found.")
|
| 440 |
-
if not ref_path.exists():
|
| 441 |
-
raise HTTPException(400, "Reference photo not found.")
|
| 442 |
-
jobs[job_id] = {"status": "queued", "progress": 0,
|
| 443 |
-
"status_text": "Queued...", "result_path": None,
|
| 444 |
-
"error": None, "stats": None, "cancelled": False, "eta": ""}
|
| 445 |
-
background_tasks.add_task(process_video, job_id, video_path, ref_path)
|
| 446 |
-
return {"job_id": job_id, "status": "queued"}
|
| 447 |
-
|
| 448 |
-
@app.get("/status/{job_id}")
|
| 449 |
-
def get_status(job_id: str):
|
| 450 |
-
if job_id not in jobs:
|
| 451 |
-
raise HTTPException(404, "Job not found")
|
| 452 |
-
j = jobs[job_id]
|
| 453 |
-
return {"status": j["status"], "progress": j["progress"],
|
| 454 |
-
"status_text": j["status_text"], "error": j.get("error"),
|
| 455 |
-
"stats": j.get("stats"), "eta": j.get("eta", "")}
|
| 456 |
-
|
| 457 |
-
@app.post("/cancel/{job_id}")
|
| 458 |
-
def cancel_job(job_id: str):
|
| 459 |
-
if job_id not in jobs:
|
| 460 |
-
raise HTTPException(404, "Job not found")
|
| 461 |
-
jobs[job_id]["cancelled"] = True
|
| 462 |
-
jobs[job_id]["status"] = "cancelled"
|
| 463 |
-
jobs[job_id]["status_text"] = "Cancelling..."
|
| 464 |
-
return {"cancelled": True}
|
| 465 |
-
|
| 466 |
-
@app.get("/preview/{job_id}")
|
| 467 |
-
def get_preview(job_id: str):
|
| 468 |
-
preview_path = TMP / job_id / "preview.jpg"
|
| 469 |
-
if not preview_path.exists():
|
| 470 |
-
raise HTTPException(404, "No preview yet")
|
| 471 |
-
return FileResponse(preview_path, media_type="image/jpeg")
|
| 472 |
-
|
| 473 |
-
@app.get("/download/{job_id}")
|
| 474 |
-
def download(job_id: str):
|
| 475 |
-
if job_id not in jobs:
|
| 476 |
-
raise HTTPException(404, "Job not found")
|
| 477 |
-
j = jobs[job_id]
|
| 478 |
-
if j["status"] != "done":
|
| 479 |
-
raise HTTPException(400, "Not complete")
|
| 480 |
-
p = Path(j["result_path"])
|
| 481 |
-
if not p.exists():
|
| 482 |
-
raise HTTPException(404, "File not found")
|
| 483 |
-
return FileResponse(p, media_type="video/mp4", filename="clipcut_result.mp4")
|
| 484 |
-
|
| 485 |
-
@app.delete("/job/{job_id}")
|
| 486 |
-
def delete_job(job_id: str):
|
| 487 |
-
d = TMP / job_id
|
| 488 |
-
if d.exists(): shutil.rmtree(d)
|
| 489 |
-
jobs.pop(job_id, None)
|
| 490 |
-
return {"deleted": True}
|
| 491 |
-
|
| 492 |
-
# ── VidFetch ───────────────────────────────────────────────────────────────────
|
| 493 |
-
|
| 494 |
-
vf_jobs: dict[str, dict] = {}
|
| 495 |
-
|
| 496 |
-
COBALT_URL = "https://runitdamar-cobalt.hf.space"
|
| 497 |
-
|
| 498 |
-
async def run_vidfetch(job_id: str, urls: list, mode: str):
|
| 499 |
-
import aiohttp
|
| 500 |
-
job = vf_jobs[job_id]
|
| 501 |
-
out_dir = TMP / job_id
|
| 502 |
-
out_dir.mkdir(parents=True, exist_ok=True)
|
| 503 |
-
downloaded = []
|
| 504 |
-
|
| 505 |
-
try:
|
| 506 |
-
total = len(urls)
|
| 507 |
-
async with aiohttp.ClientSession() as session:
|
| 508 |
-
for i, url in enumerate(urls):
|
| 509 |
-
if job.get("cancelled"): break
|
| 510 |
-
job["status_text"] = f"Fetching video {i+1} of {total}..."
|
| 511 |
-
job["progress"] = int((i / total) * 80)
|
| 512 |
-
|
| 513 |
-
# Call Cobalt API
|
| 514 |
-
try:
|
| 515 |
-
async with session.post(
|
| 516 |
-
f"{COBALT_URL}/",
|
| 517 |
-
json={"url": url},
|
| 518 |
-
headers={"Accept": "application/json", "Content-Type": "application/json"},
|
| 519 |
-
timeout=aiohttp.ClientTimeout(total=30)
|
| 520 |
-
) as resp:
|
| 521 |
-
data = await resp.json()
|
| 522 |
-
except Exception as e:
|
| 523 |
-
job["status"] = "failed"
|
| 524 |
-
job["error"] = f"Could not reach download service. Try again shortly."
|
| 525 |
-
return
|
| 526 |
-
|
| 527 |
-
# Cobalt returns either a direct URL or a tunnel URL
|
| 528 |
-
download_url = data.get("url") or data.get("tunnel")
|
| 529 |
-
if not download_url:
|
| 530 |
-
status = data.get("status", "unknown")
|
| 531 |
-
job["status"] = "failed"
|
| 532 |
-
job["error"] = f"Could not get download link for video {i+1}. The link may be private or unsupported. (Status: {status})"
|
| 533 |
-
return
|
| 534 |
-
|
| 535 |
-
# Download the actual video file
|
| 536 |
-
job["status_text"] = f"Downloading video {i+1} of {total}..."
|
| 537 |
-
job["progress"] = int((i / total) * 80) + 20
|
| 538 |
-
|
| 539 |
-
out_path = out_dir / f"video_{i:03d}.mp4"
|
| 540 |
-
try:
|
| 541 |
-
async with session.get(
|
| 542 |
-
download_url,
|
| 543 |
-
timeout=aiohttp.ClientTimeout(total=300),
|
| 544 |
-
headers={"User-Agent": "Mozilla/5.0"}
|
| 545 |
-
) as dl_resp:
|
| 546 |
-
if dl_resp.status != 200:
|
| 547 |
-
job["status"] = "failed"
|
| 548 |
-
job["error"] = f"Failed to download video {i+1}. Status: {dl_resp.status}"
|
| 549 |
-
return
|
| 550 |
-
with open(out_path, "wb") as f:
|
| 551 |
-
async for chunk in dl_resp.content.iter_chunked(1024 * 64):
|
| 552 |
-
f.write(chunk)
|
| 553 |
-
except Exception as e:
|
| 554 |
-
job["status"] = "failed"
|
| 555 |
-
job["error"] = f"Download failed for video {i+1}: {str(e)}"
|
| 556 |
-
return
|
| 557 |
-
|
| 558 |
-
if out_path.exists() and out_path.stat().st_size > 0:
|
| 559 |
-
downloaded.append(out_path)
|
| 560 |
-
else:
|
| 561 |
-
job["status"] = "failed"
|
| 562 |
-
job["error"] = f"Video {i+1} downloaded but file is empty."
|
| 563 |
-
return
|
| 564 |
-
|
| 565 |
-
if job.get("cancelled"):
|
| 566 |
-
job["status"] = "cancelled"
|
| 567 |
-
return
|
| 568 |
-
|
| 569 |
-
if not downloaded:
|
| 570 |
-
job["status"] = "failed"
|
| 571 |
-
job["error"] = "No videos were downloaded."
|
| 572 |
-
return
|
| 573 |
-
|
| 574 |
-
job["progress"] = 85
|
| 575 |
-
|
| 576 |
-
if mode == "merge" and len(downloaded) > 1:
|
| 577 |
-
job["status_text"] = "Merging videos..."
|
| 578 |
-
job["progress"] = 90
|
| 579 |
-
list_file = out_dir / "concat.txt"
|
| 580 |
-
list_file.write_text("\n".join(f"file '{p}'" for p in downloaded))
|
| 581 |
-
merged = out_dir / "merged.mp4"
|
| 582 |
-
r = subprocess.run([
|
| 583 |
-
"ffmpeg", "-y", "-f", "concat", "-safe", "0",
|
| 584 |
-
"-i", str(list_file), "-c", "copy", str(merged)
|
| 585 |
-
], capture_output=True)
|
| 586 |
-
if r.returncode != 0 or not merged.exists():
|
| 587 |
-
job["status"] = "failed"
|
| 588 |
-
job["error"] = "Failed to merge videos."
|
| 589 |
-
return
|
| 590 |
-
vi = video_info(merged)
|
| 591 |
-
job.update({
|
| 592 |
-
"status": "done", "progress": 100, "status_text": "Done!",
|
| 593 |
-
"mode": "merge", "count": len(downloaded),
|
| 594 |
-
"duration": round(vi["duration"], 1),
|
| 595 |
-
"result_path": str(merged)
|
| 596 |
-
})
|
| 597 |
-
else:
|
| 598 |
-
job.update({
|
| 599 |
-
"status": "done", "progress": 100, "status_text": "Done!",
|
| 600 |
-
"mode": "download", "count": len(downloaded),
|
| 601 |
-
"result_paths": [str(p) for p in downloaded]
|
| 602 |
-
})
|
| 603 |
-
|
| 604 |
-
except Exception as e:
|
| 605 |
-
import traceback
|
| 606 |
-
job["status"] = "failed"
|
| 607 |
-
job["error"] = f"Error: {str(e)}"
|
| 608 |
-
print(traceback.format_exc())
|
| 609 |
-
|
| 610 |
-
from pydantic import BaseModel
|
| 611 |
-
|
| 612 |
-
class VidFetchRequest(BaseModel):
|
| 613 |
-
job_id: str
|
| 614 |
-
urls: list
|
| 615 |
-
mode: str = "download"
|
| 616 |
-
|
| 617 |
-
@app.post("/vidfetch")
|
| 618 |
-
async def vidfetch(req: VidFetchRequest, background_tasks: BackgroundTasks):
|
| 619 |
-
vf_jobs[req.job_id] = {
|
| 620 |
-
"status": "queued", "progress": 0, "status_text": "Queued...",
|
| 621 |
-
"cancelled": False, "error": None
|
| 622 |
-
}
|
| 623 |
-
background_tasks.add_task(asyncio.create_task, run_vidfetch(req.job_id, req.urls, req.mode))
|
| 624 |
-
return {"job_id": req.job_id, "status": "queued"}
|
| 625 |
-
|
| 626 |
-
@app.get("/vidfetch/status/{job_id}")
|
| 627 |
-
def vidfetch_status(job_id: str):
|
| 628 |
-
if job_id not in vf_jobs:
|
| 629 |
-
raise HTTPException(404, "Job not found")
|
| 630 |
-
j = vf_jobs[job_id]
|
| 631 |
-
return {
|
| 632 |
-
"status": j["status"], "progress": j["progress"],
|
| 633 |
-
"status_text": j["status_text"], "error": j.get("error"),
|
| 634 |
-
"mode": j.get("mode"), "count": j.get("count"),
|
| 635 |
-
"duration": j.get("duration")
|
| 636 |
-
}
|
| 637 |
-
|
| 638 |
-
@app.post("/vidfetch/cancel/{job_id}")
|
| 639 |
-
def vidfetch_cancel(job_id: str):
|
| 640 |
-
if job_id not in vf_jobs:
|
| 641 |
-
raise HTTPException(404, "Job not found")
|
| 642 |
-
vf_jobs[job_id]["cancelled"] = True
|
| 643 |
-
vf_jobs[job_id]["status"] = "cancelled"
|
| 644 |
-
return {"cancelled": True}
|
| 645 |
-
|
| 646 |
-
@app.get("/vidfetch/download/{job_id}")
|
| 647 |
-
def vidfetch_download(job_id: str):
|
| 648 |
-
if job_id not in vf_jobs:
|
| 649 |
-
raise HTTPException(404, "Job not found")
|
| 650 |
-
j = vf_jobs[job_id]
|
| 651 |
-
if j["status"] != "done":
|
| 652 |
-
raise HTTPException(400, "Not complete")
|
| 653 |
-
path = Path(j.get("result_path", "") or j.get("result_paths", [""])[0])
|
| 654 |
-
if not path.exists():
|
| 655 |
-
raise HTTPException(404, "File not found")
|
| 656 |
-
return FileResponse(path, media_type="video/mp4", filename="vidix_download.mp4")
|
| 657 |
-
|
| 658 |
-
@app.get("/vidfetch/download/{job_id}/{index}")
|
| 659 |
-
def vidfetch_download_single(job_id: str, index: int):
|
| 660 |
-
if job_id not in vf_jobs:
|
| 661 |
-
raise HTTPException(404, "Job not found")
|
| 662 |
-
j = vf_jobs[job_id]
|
| 663 |
-
if j["status"] != "done":
|
| 664 |
-
raise HTTPException(400, "Not complete")
|
| 665 |
-
paths = j.get("result_paths", [])
|
| 666 |
-
if index >= len(paths):
|
| 667 |
-
raise HTTPException(404, "File not found")
|
| 668 |
-
path = Path(paths[index])
|
| 669 |
-
if not path.exists():
|
| 670 |
-
raise HTTPException(404, "File not found")
|
| 671 |
-
return FileResponse(path, media_type="video/mp4", filename=f"vidix_video_{index+1}.mp4")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|