screenshot-vision-algorithm 0.3.5__tar.gz → 0.3.10__tar.gz

{screenshot_vision_algorithm-0.3.5 → screenshot_vision_algorithm-0.3.10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: screenshot_vision_algorithm
-Version: 0.3.5
+Version: 0.3.10
 Summary: Multi-platform screenshot vision algorithms: WeChat (card bbox, speaker bands, avatar detection, OCR helpers), XHS & Zhihu (planned)
 License: MIT
 Requires-Python: >=3.10

{screenshot_vision_algorithm-0.3.5 → screenshot_vision_algorithm-0.3.10}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "screenshot_vision_algorithm"
-version = "0.3.5"
+version = "0.3.10"
 description = "Multi-platform screenshot vision algorithms: WeChat (card bbox, speaker bands, avatar detection, OCR helpers), XHS & Zhihu (planned)"
 readme = "README.md"
 license = { text = "MIT" }

{screenshot_vision_algorithm-0.3.5 → screenshot_vision_algorithm-0.3.10}/src/screenshot_vision_algorithm/android/wechat/algorithms/speaker_band.py

@@ -26,7 +26,7 @@ from screenshot_vision_algorithm.android.wechat.algorithms.template_matching imp
 CHAT_BAND_FIRST_AVATAR_TOP_GAP_REJECT_BASELINE = 30
 
 #: 首个发言人筒上移量（一个头像高度，防止几何误差），见 PRD §5 补充2。
-CHAT_FIRST_BAND_TOP_EXTEND_BASELINE = 100
+CHAT_FIRST_BAND_TOP_EXTEND_BASELINE = 20
 
 #: 底部输入区预留（与主列表 scanner 一脉，baseline）。
 CHAT_COMPOSER_RESERVE_BOTTOM_BASELINE = 160
@@ -173,7 +173,7 @@ def detect_chat_side_avatar_ytops(
     y1 = max(0, bounds.y_top)
     y2 = min(h, bounds.y_bottom_excl)
     try:
-        from processor.left_avatar_column import (
+        from screenshot_vision_algorithm.android.wechat.algorithms.avatar_column import (
             detect_left_avatar_column_layout,
             left_avatar_ytops,
         )
@@ -221,6 +221,7 @@ def build_prd_speaker_vertical_bands(
     scale_w: float,
     first_avatar_gap_reject_baseline: int = CHAT_BAND_FIRST_AVATAR_TOP_GAP_REJECT_BASELINE,
     first_band_top_extend_baseline: int = CHAT_FIRST_BAND_TOP_EXTEND_BASELINE,
+    avatar_merge_min_sep_baseline: int = 24,
 ) -> PrdChatVerticalLayout:
     """由时间序头像顶构造 PRD §5 顶隙 orphan + 左闭右开发言人筒。
 
@@ -248,13 +249,13 @@ def build_prd_speaker_vertical_bands(
 
     # 时间轴序去重：过近视为同一人头像重复检出
     merged_starts: list[int] = []
-    min_sep = int(round(24 * sw))
+    min_sep = int(round(int(avatar_merge_min_sep_baseline) * sw))
     for e in timeline:
         y = int(e.y_top)
         if not merged_starts or y - merged_starts[-1] >= min_sep:
             merged_starts.append(y)
 
-    # 首个筒上界上移（PRD §5 补充2）：上移 100px @1080，防几何误差与系统变化。
+    # 首个筒上界上移（PRD §5 补充2）：上移 20px @1080，防几何误差与系统变化。
     first_band_extend = int(round(int(first_band_top_extend_baseline) * sw))
 
     bands: list[tuple[int, int]] = []
@@ -275,6 +276,13 @@ def build_prd_speaker_vertical_bands(
         elif first_band_start < orphan_top[1]:
             orphan_top = (orphan_top[0], first_band_start)
 
+    # 无 orphan 时，首个 band 上界对齐 y_ct，保证纵向连续全覆盖
+    # （extend 较小且头像紧贴 y_ct 时可能产生小间隙）
+    if orphan_top is None and bands:
+        first_y0, first_y1 = bands[0]
+        if first_y0 > y_ct:
+            bands[0] = (y_ct, first_y1)
+
     return PrdChatVerticalLayout(
         content_bounds=bounds,
         orphan_top_band=orphan_top,

{screenshot_vision_algorithm-0.3.5 → screenshot_vision_algorithm-0.3.10}/src/screenshot_vision_algorithm/android/wechat/ocr/avatar_guard.py

@@ -19,7 +19,7 @@ except ImportError:
     cv2 = None  # type: ignore[assignment]
 
 try:
-    from processor.left_avatar_column import AvatarCentroid
+    from screenshot_vision_algorithm.android.wechat.algorithms.avatar_column import AvatarCentroid
 except ImportError:
     AvatarCentroid = None  # type: ignore[misc, assignment]
 

{screenshot_vision_algorithm-0.3.5 → screenshot_vision_algorithm-0.3.10}/src/screenshot_vision_algorithm/android/wechat/ocr/nickname_binding.py

@@ -230,6 +230,32 @@ class NicknameOcrConfig:
     """Prefix / substring tags that exclude a block from nickname
     consideration even if all geometric criteria pass."""
 
+    # ── 以下字段可由 rules_config.yaml 覆盖（housekeeping 胶水层读 yaml → 灌参）──
+
+    skip_block_patterns: tuple[str, ...] = ()
+    """附加正则列表，逐条编译后传递给 ``_should_skip_block``。
+    内置正则（时间戳/纯数字/系统提示等）**始终并行生效**，
+    config 提供的是追加规则，不替代内置逻辑。"""
+
+    inside_y_tolerance_px: float = 4.0
+    """``_completely_inside`` 的 Y 方向容差（px @ 1080 基准 × scale_w）。
+    默认 4.0 与 ``card_binding.inside_y_tolerance_px`` 对齐。"""
+
+    cross_band_distance_threshold_px: float = 50.0
+    """Step C 跨 band 距离门控基线（px @ 1080 × scale_w）。
+    文字底部到卡片所属 speaker_band 顶部超过此值则不作为昵称。"""
+
+    # ── speaker_band 参数（同名传给 build_prd_speaker_vertical_bands）──
+
+    first_avatar_gap_reject_baseline: int = 30
+    """orphan 间隙阈值（px @ 1080）。"""
+
+    first_band_top_extend_baseline: int = 20
+    """首个 band 上移量（px @ 1080）。"""
+
+    avatar_merge_min_sep_px: int = 24
+    """头像合并最小间距（px @ 1080 × scale_w）。"""
+
 
 _ATTRIBUTION_PREFIX_DROPS: tuple[str, ...] = (
     "撤回了一条消息",
@@ -351,6 +377,9 @@ _PURE_TIME_RE = re.compile(r"^\d{1,2}:\d{2}(:\d{2})?$")
 _WEEKDAY_TIME_RE = re.compile(
     r"^周[一二三四五六日天].*?(?:上午|下午|晚上|凌晨)?\s*\d{0,2}[:：]?\d{0,2}"
 )
+_YESTERDAY_TODAY_TIME_RE = re.compile(
+    r"^(?:昨天|今天)\s*\d{0,2}:?\d{0,2}$"
+)
 _MOBILE_PHONE_SUFFIX_RE = re.compile(r"1\d{10}\s*$")
 
 
@@ -388,6 +417,8 @@ def _is_chat_timestamp_or_divider(text: str) -> bool:
         return True
     if _WEEKDAY_TIME_RE.match(t):
         return True
+    if _YESTERDAY_TODAY_TIME_RE.match(t):
+        return True
     return False
 
 
@@ -411,7 +442,7 @@ def _is_avatar_column_nickname_row(
     avatar_layout: Any = None,
 ) -> bool:
     try:
-        from processor.left_avatar_column import nickname_bbox_in_avatar_column
+        from screenshot_vision_algorithm.android.wechat.algorithms.avatar_column import nickname_bbox_in_avatar_column
     except ImportError:
         nickname_bbox_in_avatar_column = None  # type: ignore[assignment,misc]
     if avatar_layout is not None and nickname_bbox_in_avatar_column is not None:
@@ -446,7 +477,7 @@ def _avatar_anchor_for_block(
     if avatar_layout is None or getattr(avatar_layout, "empty", True):
         return None
     try:
-        from processor.left_avatar_column import find_avatar_anchor_for_nickname_bbox
+        from screenshot_vision_algorithm.android.wechat.algorithms.avatar_column import find_avatar_anchor_for_nickname_bbox
     except ImportError:
         return None
     return find_avatar_anchor_for_nickname_bbox(
@@ -463,9 +494,9 @@ def _load_left_avatar_layout(
     if png_path is None:
         return None
     try:
-        from processor.chat_bubble_geometry_prd import compute_chat_content_vertical_bounds
-        from processor.left_avatar_column import detect_left_avatar_column_layout
-        from processor.nickname_avatar_guard import load_png_bgr
+        from screenshot_vision_algorithm.android.wechat.algorithms.speaker_band import compute_chat_content_vertical_bounds
+        from screenshot_vision_algorithm.android.wechat.algorithms.avatar_column import detect_left_avatar_column_layout
+        from screenshot_vision_algorithm.android.wechat.ocr.avatar_guard import load_png_bgr
     except ImportError:
         return None
     arr = load_png_bgr(str(png_path))
@@ -595,7 +626,7 @@ def _nickname_passes_avatar_guard(
     if bgr_guard is None:
         return not config.require_avatar_guard_for_nickname
     try:
-        from processor.nickname_avatar_guard import avatar_roi_pass
+        from screenshot_vision_algorithm.android.wechat.ocr.avatar_guard import avatar_roi_pass
     except ImportError:
         return not config.require_avatar_guard_for_nickname
 
@@ -687,13 +718,13 @@ def _load_prd_layout_optional(
     if png_path is None or not config.use_prd_chat_vertical_bands:
         return None
     try:
-        from processor.chat_bubble_geometry_prd import (
+        from screenshot_vision_algorithm.android.wechat.algorithms.speaker_band import (
             build_prd_speaker_vertical_bands,
             compute_chat_content_vertical_bounds,
             detect_chat_side_avatar_ytops,
             merge_avatar_ytops_time_order,
         )
-        from processor.nickname_avatar_guard import load_png_bgr
+        from screenshot_vision_algorithm.android.wechat.ocr.avatar_guard import load_png_bgr
     except ImportError:
         return None
     arr = load_png_bgr(str(png_path))
@@ -705,7 +736,12 @@ def _load_prd_layout_optional(
     bounds = compute_chat_content_vertical_bounds(arr, sw, screen_h=oh)
     ly, ry = detect_chat_side_avatar_ytops(arr, sw, bounds)
     timeline = merge_avatar_ytops_time_order(ly, ry)
-    return build_prd_speaker_vertical_bands(timeline, bounds, scale_w=sw)
+    return build_prd_speaker_vertical_bands(
+        timeline, bounds, scale_w=sw,
+        first_avatar_gap_reject_baseline=config.first_avatar_gap_reject_baseline,
+        first_band_top_extend_baseline=config.first_band_top_extend_baseline,
+        avatar_merge_min_sep_baseline=config.avatar_merge_min_sep_px,
+    )
 
 
 def _finalize_prd_attribution(
@@ -882,7 +918,7 @@ def _resume_thumb_bindings_and_orphans(
     # OCR bbox 与卡片 bbox 各有 ±2~3px 抖动：卡内标题行可能比卡顶高出 1~3px，
     # 若按严格包含判定会被误认为"卡外文字"而误绑为昵称（2026-06-11 真实案例）。
     # 真昵称行整体位于卡片上方（y2 < 卡顶），不受此容差影响。
-    _inside_tol = 4.0 * max(scale_w, 1.0)
+    _inside_tol = float(config.inside_y_tolerance_px) * max(scale_w, 1.0)
 
     def _completely_inside(block_bbox: tuple, card_bbox: tuple) -> bool:
         bx1, by1, bx2, by2 = block_bbox
@@ -912,8 +948,25 @@ def _resume_thumb_bindings_and_orphans(
             return True
         if _is_favorite_card_label(t):
             return True
+        # ── 附加正则（来自 rules_config.yaml skip_patterns，编译后缓存）──
+        if _compiled_skip_patterns:
+            for _pat in _compiled_skip_patterns:
+                try:
+                    if _pat.search(t):
+                        return True
+                except Exception:
+                    continue
         return False
 
+    # ── 编译 config 追加的正则（函数内缓存，避免每次 block 重新编译）──
+    _compiled_skip_patterns: list[re.Pattern] = []
+    if config.skip_block_patterns:
+        for p in config.skip_block_patterns:
+            try:
+                _compiled_skip_patterns.append(re.compile(p))
+            except re.error:
+                pass
+
     bindings: list[ResumeThumbBinding] = []
     orphan = 0
     excluded_indices: set[int] = set()
@@ -962,7 +1015,17 @@ def _resume_thumb_bindings_and_orphans(
             _cy1, _cy2 = float(tbb[1]), float(tbb[3])
             for _bi, (_b0, _b1) in enumerate(_sb_list):
                 if _cy2 > float(_b0) and _cy1 < float(_b1):
-                    band_idx = _bi  # 相交 → 候选；持续覆盖 → 最后一个（下方）胜出
+                        band_idx = _bi  # 相交 → 候选；持续覆盖 → 最后一个（下方）胜出
+
+        # ── 如果卡片与 orphan_top_band 相交 → 直接记为找不到归属 ──
+        if layout_opt is not None:
+            ot = getattr(layout_opt, "orphan_top_band", None)
+            if ot is not None:
+                oy0, oy1 = float(ot[0]), float(ot[1])
+                _cy1, _cy2 = float(tbb[1]), float(tbb[3])
+                if _cy2 > oy0 and _cy1 < oy1:
+                    orphan += 1
+                    continue
 
         matched_nickname: Optional[str] = None
 
@@ -1053,14 +1116,14 @@ def _resume_thumb_bindings_and_orphans(
 
             if best_i is not None:
                 # PRD §B.1 C：距离检查——文字底部到卡片所属 speaker_band 顶部
-                # 超过阈值（50px baseline × scale_w）则不作为昵称
+                # 超过阈值（config.cross_band_distance_threshold_px × scale_w）则不作为昵称
                 _use_as_nickname = True
                 if band_idx is not None and band_idx >= 0:
                     _sb_list = getattr(layout_opt, "speaker_bands", ())
                     if band_idx < len(_sb_list):
                         _band_y0 = float(_sb_list[band_idx][0])
                         _gap = _band_y0 - best_bottom  # positive → block above band
-                        _threshold = 50.0 * float(scale_w)
+                        _threshold = float(config.cross_band_distance_threshold_px) * float(scale_w)
                         if _gap > _threshold:
                             _use_as_nickname = False
                 if _use_as_nickname:
@@ -1177,7 +1240,7 @@ def first_attrib_verbatim_display_line_and_bbox(
     scale_w = float(original_width) / 1080.0
     if png_path is not None:
         try:
-            from processor.nickname_avatar_guard import load_png_bgr
+            from screenshot_vision_algorithm.android.wechat.ocr.avatar_guard import load_png_bgr
 
             bgr_guard = load_png_bgr(str(png_path))
         except ImportError:
@@ -1748,7 +1811,7 @@ def extract_nicknames(
     bgr_guard = None
     if png_path is not None:
         try:
-            from processor.nickname_avatar_guard import load_png_bgr
+            from screenshot_vision_algorithm.android.wechat.ocr.avatar_guard import load_png_bgr
 
             bgr_guard = load_png_bgr(str(png_path))
         except ImportError:

{screenshot_vision_algorithm-0.3.5 → screenshot_vision_algorithm-0.3.10}/src/screenshot_vision_algorithm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: screenshot_vision_algorithm
-Version: 0.3.5
+Version: 0.3.10
 Summary: Multi-platform screenshot vision algorithms: WeChat (card bbox, speaker bands, avatar detection, OCR helpers), XHS & Zhihu (planned)
 License: MIT
 Requires-Python: >=3.10