screenshot-vision-algorithm 0.3.1__tar.gz → 0.3.2__tar.gz

{screenshot_vision_algorithm-0.3.1 → screenshot_vision_algorithm-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: screenshot_vision_algorithm
-Version: 0.3.1
+Version: 0.3.2
 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-0.3.1 → screenshot_vision_algorithm-0.3.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "screenshot_vision_algorithm"
-version = "0.3.1"
+version = "0.3.2"
 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-0.3.1 → screenshot_vision_algorithm-0.3.2}/src/screenshot_vision_algorithm/__init__.py

@@ -3,7 +3,6 @@
 from screenshot_vision_algorithm._config import Platform, WeChatVersion, Profile
 
 from screenshot_vision_algorithm.android.wechat.algorithms.card_bbox import (
-    TAP_ANTI_DETECT_X_JITTER_PX,
     _compute_exact_card_and_bubble_bboxes,
     _compute_exact_card_bboxes,
     BubbleBbox,

{screenshot_vision_algorithm-0.3.1 → screenshot_vision_algorithm-0.3.2}/src/screenshot_vision_algorithm/android/wechat/algorithms/card_bbox.py

@@ -83,11 +83,6 @@ REFERENCE_RESUME_CARD_TOP_GAP_BASELINE: int = 542
 #: **相对 screencap + OpenCV**，性能可忽略。
 PROCESSED_CARD_MATCH_MIN_OVERLAP_RATIO: float = 0.70
 
-#: Anti-detection horizontal jitter (px @ 1080 baseline) applied by
-#: ``enter_note`` on every tap attempt: ``tap_x = click_x ± random(0..15)``.
-#: Y is never jittered (speaker attribution depends on exact Y).
-TAP_ANTI_DETECT_X_JITTER_PX: int = 15
-
 #: 3b gap 合并/延长判别距离（px @ 1080 baseline，运行时 ×scale_w）：
 #: 收藏锚点落在同一竖线相邻两 zone 的间隙时，取锚点下方第一条横线，
 #: 横线与下方 zone 起点距离 ≤ 本值 → 两 zone 同卡（被收藏标签行打断），合并；
@@ -502,7 +497,7 @@ def _compute_exact_card_and_bubble_bboxes(
             if exact_y2 <= exact_y1 or exact_y2 - exact_y1 < 80:
                 continue
 
-        click_x = min(screen_w - 1, max(0, hit.x + hit.w // 2))
+        click_x = screen_w // 2
         click_y = min(screen_h - 1, max(0, hit.y + hit.h // 2))
         midline = screen_w / 2.0
         click_side = "left" if click_x < midline else "right"
@@ -648,10 +643,11 @@ def derive_cards(
     ordered = sorted(hits, key=lambda h: h.y)
 
     # New algorithm: vline-segment + hline-boundary (PRD S6). When the
-    # pipeline ran successfully its verdict is final — an anchor it dropped
-    # (no containing zone → "点对卡、记错框" guard) must NOT be resurrected
-    # by the legacy fixed-offset fallback. Legacy only serves chat_img=None
-    # callers and pipeline exceptions.
+    # pipeline ran successfully with non-empty cards its verdict is final —
+    # an anchor it dropped (no containing zone → "点对卡、记错框" guard)
+    # must NOT be resurrected by the legacy fixed-offset fallback. When the
+    # pipeline produces no cards (e.g. variance texture insufficient),
+    # legacy fixed-offset serves as a reasonable fallback.
     if chat_img is not None and len(ordered) > 0:
         try:
             exact_cards = _compute_exact_card_bboxes(
@@ -660,9 +656,11 @@ def derive_cards(
         except Exception:
             pass
         else:
-            return drop_top_clamped_false_positive_cards(
-                exact_cards, scale_w=scale_w,
-            )
+            if exact_cards:
+                return drop_top_clamped_false_positive_cards(
+                    exact_cards, scale_w=scale_w,
+                )
+            # exact_cards empty → fall through to legacy
 
     # Legacy: fixed-offset algorithm
     top_offset = int(round(FAVORITE_TO_CARD_TOP_OFFSET_BASELINE * scale_w))
@@ -681,7 +679,7 @@ def derive_cards(
         if card_bottom <= card_top:
             continue
 
-        click_x = min(screen_w - 1, max(0, hit.x + hit.w // 2))
+        click_x = screen_w // 2
         click_y = min(screen_h - 1, max(0, hit.y + hit.h // 2))
         midline = screen_w / 2
         click_side = "left" if click_x < midline else "right"
@@ -979,7 +977,6 @@ __all__ = [
     "GAP_HLINE_TO_NEXT_ZONE_MAX_BASELINE",
     "PROCESSED_CARD_MATCH_MIN_OVERLAP_RATIO",
     "REFERENCE_RESUME_CARD_TOP_GAP_BASELINE",
-    "TAP_ANTI_DETECT_X_JITTER_PX",
     "FavoriteLabelHit",
     "ThumbnailCard",
     "TrackedCard",

{screenshot_vision_algorithm-0.3.1 → screenshot_vision_algorithm-0.3.2}/src/screenshot_vision_algorithm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: screenshot_vision_algorithm
-Version: 0.3.1
+Version: 0.3.2
 Summary: Multi-platform screenshot vision algorithms: WeChat (card bbox, speaker bands, avatar detection, OCR helpers), XHS & Zhihu (planned)
 License: MIT
 Requires-Python: >=3.10