Fix S3 streaming upload + detect white/gray watermarks

S3 upload:
- Replace read_bytes() (loads entire file into memory) with streaming SHA256
  computation (8MB chunks) + httpx streaming PUT with Content-Length header
- Add detailed logging for each upload step to diagnose future failures

Watermark detection (smart_detect.py):
- Add Path B: brightness-diff detection for low-saturation watermarks like
  white/gray text (e.g. 'sjhs.pw'). Compares mean brightness in first 10s
  vs the rest of the video. Regions that are brighter in early frames AND
  located in frame corners (top-left, top-right, etc.) are flagged.
- Merge Path A (saturation-based) + Path B (brightness-diff) results
- Fall back to global std method only if both paths find nothing

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

backend/tools/smart_detect.py

@@ -76,51 +76,58 @@ def detect_static_regions(video_path: str, sample_frames: int = 30, threshold: f
     # 分析时使用小尺寸坐标空间
     w, h = SW, SH
 
-    if len(logo_gray_frames) < 2:
-        print(f"[SAM2-detect] logo帧不足，回退到时序std法", flush=True)
+    # ── 路径A：彩色水印（高饱和度）检测 ──────────────────────────────────
+    sat_mask = np.zeros((h, w), dtype=np.uint8)
+    if len(logo_gray_frames) >= 2 and nlogo_bgr_frames:
+        sat_acc = np.zeros((h, w), dtype=np.float32)
+        for bgr in logo_bgr_frames:
+            sat_acc += cv2.cvtColor(bgr, cv2.COLOR_BGR2HSV)[:, :, 1].astype(np.float32) / 255.0
+        sat_mean_logo = sat_acc / len(logo_bgr_frames)
+
+        sat_acc_nl = np.zeros((h, w), dtype=np.float32)
+        for bgr in nlogo_bgr_frames[:6]:
+            sat_acc_nl += cv2.cvtColor(bgr, cv2.COLOR_BGR2HSV)[:, :, 1].astype(np.float32) / 255.0
+        sat_mean_nlogo = sat_acc_nl / min(len(nlogo_bgr_frames), 6)
+        sat_diff = sat_mean_logo - sat_mean_nlogo
+
+        sat_thresh = min(0.6, np.percentile(sat_mean_logo, 92))
+        diff_thresh = max(0.25, sat_diff.mean() + sat_diff.std())
+        sat_mask = ((sat_mean_logo >= sat_thresh) & (sat_diff >= diff_thresh)).astype(np.uint8) * 255
+        print(f"[SAM2-detect] 彩色路径: sat_thresh={sat_thresh:.3f} diff_thresh={diff_thresh:.3f} 候选={np.sum(sat_mask>0)}", flush=True)
+
+    # ── 路径B：白色/灰色水印（亮度差分）检测 ────────────────────────────────
+    # "sjhs.pw" 类白色文字水印：饱和度≈0，但亮度比背景高，且只出现在少数帧
+    # 策略：把前 N 秒的帧与后续帧做亮度差分，找只出现在片头的高亮区域
+    bright_mask = np.zeros((h, w), dtype=np.uint8)
+    early_cutoff = min(len(gray_frames), max(5, int(10 * fps / step)))  # 前10秒
+    if early_cutoff >= 2 and len(gray_frames) > early_cutoff:
+        early_stack = np.stack(gray_frames[:early_cutoff], axis=0)
+        rest_stack  = np.stack(gray_frames[early_cutoff:], axis=0)
+        early_mean = early_stack.mean(axis=0)
+        rest_mean  = rest_stack.mean(axis=0)
+        # 片头比后续亮得多 → 水印像素（值越大越亮）
+        bright_diff = early_mean - rest_mean
+        # 阈值：差值超过 20 灰度级的区域
+        bright_mask_raw = (bright_diff > 20).astype(np.uint8) * 255
+        # 只保留四角区域（水印通常在角落），避免误报视频内容
+        corner_mask = np.zeros((h, w), dtype=np.uint8)
+        ch, cw = h // 4, w // 4   # 角落占画面 25%
+        corner_mask[:ch, :cw] = 1          # 左上
+        corner_mask[:ch, -cw:] = 1         # 右上
+        corner_mask[-ch:, :cw] = 1         # 左下
+        corner_mask[-ch:, -cw:] = 1        # 右下
+        bright_mask = cv2.bitwise_and(bright_mask_raw, corner_mask * 255)
+        print(f"[SAM2-detect] 亮度差分路径: bright_diff_max={bright_diff.max():.1f} 角落候选={np.sum(bright_mask>0)}", flush=True)
+
+    # ── 合并两条路径 ──────────────────────────────────────────────────────
+    static_mask = cv2.bitwise_or(sat_mask, bright_mask)
+    if np.sum(static_mask) == 0:
+        # 两条路径都没找到，回退到全局时序std法
+        print(f"[SAM2-detect] 两条路径均无结果，回退到时序std法", flush=True)
         stack = np.stack(gray_frames, axis=0)
         std_map = np.std(stack, axis=0)
         static_mask = (std_map < 8.0).astype(np.uint8) * 255
-    else:
-        # 在logo帧中统计每像素平均饱和度
-        sat_acc = np.zeros((h, w), dtype=np.float32)
-        for bgr in logo_bgr_frames:
-            hsv = cv2.cvtColor(bgr, cv2.COLOR_BGR2HSV)
-            sat_acc += (hsv[:, :, 1].astype(np.float32) / 255.0)
-        sat_mean_logo = sat_acc / len(logo_bgr_frames)
-
-        if nlogo_bgr_frames:
-            # ── 有无logo帧对比：准确检测logo区域 ───────────────────────────
-            # Logo像素: logo帧高饱和 + 无logo帧低饱和（差值大）
-            sat_acc_nl = np.zeros((h, w), dtype=np.float32)
-            for bgr in nlogo_bgr_frames[:6]:
-                hsv = cv2.cvtColor(bgr, cv2.COLOR_BGR2HSV)
-                sat_acc_nl += (hsv[:, :, 1].astype(np.float32) / 255.0)
-            sat_mean_nlogo = sat_acc_nl / min(len(nlogo_bgr_frames), 6)
-            sat_diff = sat_mean_logo - sat_mean_nlogo
-
-            sat_thresh = min(0.6, np.percentile(sat_mean_logo, 92))
-            diff_thresh = max(0.25, sat_diff.mean() + sat_diff.std())
-            logo_pixel_mask = ((sat_mean_logo >= sat_thresh) & (sat_diff >= diff_thresh)).astype(np.uint8) * 255
-
-            print(f"[SAM2-detect] logo帧平均饱和: max={sat_mean_logo.max():.3f}", flush=True)
-            print(f"[SAM2-detect] 饱和差值: max={sat_diff.max():.3f} mean={sat_diff.mean():.3f}", flush=True)
-            print(f"[SAM2-detect] sat_thresh={sat_thresh:.3f} diff_thresh={diff_thresh:.3f}, 候选={np.sum(logo_pixel_mask > 0)}", flush=True)
-        else:
-            # ── 无非logo帧（短视频/预览：logo全程可见）──────────────────────
-            # 只在帧的边缘区域（顶部30%）寻找高饱和logo
-            # 避免把视频中央的自然内容误报为logo
-            border_mask = np.zeros((h, w), dtype=np.uint8)
-            border_mask[:h * 3 // 10, :] = 1  # 顶部 30%
-
-            # 取最高饱和的4%像素（极鲜艳 = logo文字/图形）
-            sat_thresh = max(0.6, np.percentile(sat_mean_logo, 96))
-            logo_pixel_mask = ((sat_mean_logo >= sat_thresh).astype(np.uint8) * border_mask * 255)
-
-            print(f"[SAM2-detect] 无非logo帧，限定边缘区域搜索", flush=True)
-            print(f"[SAM2-detect] sat_thresh={sat_thresh:.3f}, 候选={np.sum(logo_pixel_mask > 0)}", flush=True)
-
-        static_mask = logo_pixel_mask
+    print(f"[SAM2-detect] 合并后候选像素数={np.sum(static_mask>0)}", flush=True)
 
     # 仅排除纯黑区域
     mean_gray = np.mean(np.stack(gray_frames, axis=0), axis=0)

web_app.py

@@ -834,17 +834,26 @@ class VSRTaskWorker:
         logger.info(f"[vsr_worker] 任务 #{platform_task_id} 上报结果 status={status}")
 
     async def _upload_result_video(self, local_path: str) -> str | None:
-        """处理完成后将输出视频上传到平台，返回 play_url（S3 地址）。"""
+        """处理完成后将输出视频上传到平台，返回 play_url（S3 地址）。流式上传，不一次性读入内存。"""
         import hashlib as _hl
         try:
-            data = Path(local_path).read_bytes()
-            sha256 = _hl.sha256(data).hexdigest()
-            filename = Path(local_path).name
+            file_path = Path(local_path)
+            file_size = file_path.stat().st_size
+            filename = file_path.name
+
+            # 流式计算 SHA256，避免 read_bytes() 把几百 MB 全加载进内存
+            h = _hl.sha256()
+            with open(local_path, "rb") as f:
+                for chunk in iter(lambda: f.read(8 * 1024 * 1024), b""):
+                    h.update(chunk)
+            sha256 = h.hexdigest()
+            logger.info(f"[vsr_worker] 文件 {filename} size={file_size//1048576}MB sha256={sha256[:12]}…")
+
             sign_resp = await self._signed_post(
                 "/openapi/file/upload/get_upload_sign",
                 {
                     "sha256": sha256,
-                    "size": len(data),
+                    "size": file_size,
                     "filename": filename,
                     "content_type": "video/mp4",
                     "file_type": "video",
@@ -853,24 +862,36 @@ class VSRTaskWorker:
                 },
             )
             if sign_resp is None:
+                logger.warning("[vsr_worker] get_upload_sign 返回空，跳过上传")
                 return None
             action = sign_resp.get("action", "")
             if action == "skip":
+                logger.info(f"[vsr_worker] 文件已存在，直接返回 play_url")
                 return sign_resp.get("play_url")
             presigned_url = sign_resp.get("presigned_url", "")
             object_key = sign_resp.get("object_key", "")
             if not presigned_url or not object_key:
                 logger.warning(f"[vsr_worker] upload sign 未返回 presigned_url/object_key: {sign_resp}")
                 return None
-            put_resp = await self._client.put(presigned_url, content=data,
-                                              headers={"Content-Type": "video/mp4"})
+
+            # 流式 PUT 上传，不把文件加载到内存
+            logger.info(f"[vsr_worker] 开始流式上传 {file_size//1048576}MB → S3…")
+            async with httpx.AsyncClient(timeout=7200) as upload_client:
+                with open(local_path, "rb") as f:
+                    put_resp = await upload_client.put(
+                        presigned_url,
+                        content=f,
+                        headers={"Content-Type": "video/mp4", "Content-Length": str(file_size)},
+                    )
             put_resp.raise_for_status()
+            logger.info(f"[vsr_worker] S3 上传完成，注册资产…")
+
             reg_resp = await self._signed_post(
                 "/openapi/file/assets/register",
                 {
                     "sha256": sha256,
                     "object_key": object_key,
-                    "size": len(data),
+                    "size": file_size,
                     "filename": filename,
                     "content_type": "video/mp4",
                     "file_type": "video",
@@ -878,7 +899,9 @@ class VSRTaskWorker:
                     "no_transcode": True,
                 },
             )
-            return reg_resp.get("play_url") if reg_resp else None
+            play_url = reg_resp.get("play_url") if reg_resp else None
+            logger.info(f"[vsr_worker] 资产注册完成 play_url={play_url}")
+            return play_url
         except Exception as e:
             logger.warning(f"[vsr_worker] 上传结果视频失败: {e}")
             return None