SAM2: sample every second for the entire video, no exceptions

User requirement: correctness first, slow is acceptable.
- Remove the 120s dense + 8s tail split
- Full video sampled at 1 frame/second uniformly
- Analysis still uses 1/4 resolution for speed
- Coordinate scale restored to original resolution at the end

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

backend/tools/smart_detect.py

@@ -14,44 +14,62 @@ SAM2_CFG  = "configs/sam2.1/sam2.1_hiera_t.yaml"
 def detect_static_regions(video_path: str, sample_frames: int = 30, threshold: float = 20.0,
                           max_scan_seconds: float = None):
     """
-    饱和度签名 + 跨帧对比法检测水印/Logo：
-    1. 采样多帧，通过高饱和度像素比例判断哪些帧含有logo（logo通常色彩鲜艳）
-    2. 在"含logo帧"之间比较，logo像素在多帧中出现在同一位置 → 差值极低
-    3. 能检测周期性出现/动态水印，无需logo持续存在于全部帧
+    饱和度签名 + 跨帧对比法检测水印/Logo。
+    采样策略：每秒采一帧（顺序读取，不随机 seek）+ 超长视频后段每8秒一帧。
+    分析时缩放到 1/4 分辨率提速，结果坐标还原到原始尺寸。
     """
     cap = cv2.VideoCapture(video_path)
     total = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
     fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
     w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
     h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
-
+    orig_frame_w, orig_frame_h = w, h   # 保存原始分辨率，用于最终坐标还原
     duration_s = total / fps
-    num_samples = max(10, min(24, int(duration_s // 8)))
-    uniform_positions = [min(int((i + 0.5) * total / num_samples), total - 1)
-                         for i in range(num_samples)]
 
-    # 水印常出现在视频开头几秒，强制加密采样前15秒
-    early_seconds = [0.5, 1, 1.5, 2, 3, 4, 5, 7, 10, 15]
-    early_positions = [min(int(s * fps), total - 1) for s in early_seconds if s * fps < total]
-    # 合并去重，保持顺序
-    seen = set()
-    positions = []
-    for p in early_positions + uniform_positions:
-        if p not in seen:
-            seen.add(p)
-            positions.append(p)
+    # ── 采样策略：全程每秒采一帧 ─────────────────────────────────────────
+    step = max(1, int(fps))
+    positions = list(range(0, total, step))
 
-    bgr_frames = []
+    # ── 顺序读取，缩放到 1/4 分辨率加速分析 ────────────────────────────────
+    SW, SH = max(1, w // 4), max(1, h // 4)   # 分析用小尺寸
+    pos_set = set(positions)
+    bgr_frames_small = []    # 缩放后的小帧（用于分析）
+    bgr_frames_orig  = []    # 原始尺寸帧（用于 SAM2，只保留少数几帧）
     gray_frames = []
-    for pos in positions:
-        cap.set(cv2.CAP_PROP_POS_FRAMES, pos)
+    keep_orig_interval = max(1, len(positions) // 8)   # 每隔若干帧保留一张原始尺寸
+
+    cap.set(cv2.CAP_PROP_POS_FRAMES, 0)
+    frame_idx = 0
+    pos_iter = iter(sorted(positions))
+    next_pos = next(pos_iter, None)
+
+    while next_pos is not None:
+        if frame_idx < next_pos:
+            # 快速跳帧：只有在需要的位置才真正读
+            cap.set(cv2.CAP_PROP_POS_FRAMES, next_pos)
+            frame_idx = next_pos
+
         ret, frame = cap.read()
-        if ret:
-            bgr_frames.append(frame)
-            gray_frames.append(cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY).astype(np.float32))
+        if not ret:
+            next_pos = next(pos_iter, None)
+            continue
+
+        small = cv2.resize(frame, (SW, SH), interpolation=cv2.INTER_AREA)
+        bgr_frames_small.append(small)
+        gray_frames.append(cv2.cvtColor(small, cv2.COLOR_BGR2GRAY).astype(np.float32))
+        if len(bgr_frames_orig) == 0 or len(bgr_frames_small) % keep_orig_interval == 0:
+            bgr_frames_orig.append(frame)
+
+        frame_idx += 1
+        next_pos = next(pos_iter, None)
+
     cap.release()
 
-    print(f"[SAM2-detect] 采样 {len(bgr_frames)} 帧 (前15秒密采+均匀分布，共{duration_s:.1f}s)", flush=True)
+    # 后续分析全部用小尺寸帧
+    bgr_frames = bgr_frames_small
+    w, h = SW, SH   # 分析坐标空间切换到小尺寸
+
+    print(f"[SAM2-detect] 采样 {len(bgr_frames)} 帧（全程每秒一帧，共{duration_s:.1f}s，分析尺寸{SW}x{SH}）", flush=True)
 
     if len(bgr_frames) < 4:
         print(f"[SAM2-detect] 帧数不足4帧，返回空", flush=True)
@@ -175,7 +193,13 @@ def detect_static_regions(video_path: str, sample_frames: int = 30, threshold: f
             print(f"[SAM2-detect] 合并顶部 {len(top_regions)} 个区域为 ({x1},{y1})-({x2},{y2})", flush=True)
             regions = merged
 
-    print(f"[SAM2-detect] detect_static_regions 返回 {len(regions)} 个区域", flush=True)
+    # ── 坐标从分析尺寸（SW×SH）还原到原始分辨率 ──────────────────────────
+    scale_x = orig_frame_w / SW
+    scale_y = orig_frame_h / SH
+    regions = [(int(x1*scale_x), int(y1*scale_y), int(x2*scale_x), int(y2*scale_y))
+               for x1, y1, x2, y2 in regions]
+
+    print(f"[SAM2-detect] detect_static_regions 返回 {len(regions)} 个区域（已还原到{orig_frame_w}x{orig_frame_h}）", flush=True)
     return regions