Merge:'fix/RedDot-Log&v2'| 二值化红点v2识别 & 补全maa回调。#299
Open
sunyink wants to merge 5 commits into
Open
Conversation
Contributor
📝 WalkthroughWalkthroughRedDotDetector 从绝对阈值迁移为分项归一化置信度加权（min_confidence），重写预设/独立执行路径、异常返回为结构化 AnalyzeResult、新增置信度/诊断计算与按 key 节流的调试落盘功能。 ChangesRedDotDetector 置信度加权模型重设计
🎯 4 (Complex) | ⏱️ ~60 minutes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
审查者指南重构 RedDotDetector 识别流水线，改为使用带置信度评分的感叹号检测器，并增加更丰富的日志/遥测与自动调试图像导出，同时增强预设/独立模式的交互和错误处理。 文件级变更
提示与命令与 Sourcery 交互
自定义你的体验访问你的 控制面板 以：
获取帮助Original review guide in EnglishReviewer's GuideRefactors the RedDotDetector recognition pipeline to use a confidence-scored exclamation detector with richer logging/telemetry and automatic debug image dumping, and enhances preset/standalone interaction and error handling. File-Level Changes
Tips and commandsInteracting with Sourcery
Customizing Your ExperienceAccess your dashboard to:
Getting Help
|
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
assets/resource/base/pipeline/Test.json (1)
668-714:⚠️ Potential issue | 🟠 Major | ⚡ Quick win清理仍在使用的 RedDotDetector 旧参数：
Agt_RedDot_Preset仍包含inner_v_min/inner_s_max
在assets/resource/base/pipeline/Dummy.json里，Agt_RedDot_Preset的custom_recognition_param仍使用了inner_v_min/inner_s_max（位于预设参数中）。而Test.json的RedDot_Preset已按 v2 模型迁移到min_confidence，文档明确该模式“不再依赖 removed inner_v_min / inner_s_max thresholds”。当前多处节点仍引用{"preset":"Agt_RedDot_Preset"}（如QuestList.json、Pass.json、Mail.json、GACHA.json、GACHA_ADV.json等），因此这些节点的红点检测很可能不会按 v2 期望的门限生效；需要把Agt_RedDot_Preset也迁移到min_confidence模型并统一参数集。🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@assets/resource/base/pipeline/Test.json` around lines 668 - 714, The Agt_RedDot_Preset still uses removed thresholds inner_v_min/inner_s_max while RedDot_Preset has been migrated to the v2 style (min_confidence, gap_ratio, red_area, hsv_ranges, debug); update the Agt_RedDot_Preset definition to the same v2 parameter set (replace inner_v_min/inner_s_max with min_confidence and include gap_ratio, red_area, hsv_ranges, debug as in RedDot_Preset) and then verify/adjust every node that references {"preset":"Agt_RedDot_Preset"} (e.g., nodes similar to CheckPanel_A) to either point to the updated Agt_RedDot_Preset or to RedDot_Preset so all red-dot detections use the unified min_confidence-based thresholds.
🧹 Nitpick comments (1)
agent/recognition/binarymatch.py (1)
398-403: ⚡ Quick win示例里的
CheckPanel_A.roi还是旧值。这里的示例 B 仍然写着
[640, 616, 17, 16]，但这次 PR 引用的配置说明已经把CheckPanel_A的 ROI 更新成[723, 616, 18, 18]。这段注释本身就是调参文档，继续保留旧坐标会把后续对照示例的人直接带偏。建议同步成当前示例值，或至少明确标注这是占位示例。As per coding guidelines, "CheckPanel_A: roi changed to [723, 616, 18, 18]; ensure this ROI matches the expected red-dot search region since the detector now crops ROI and applies the new internal blob/confidence logic."
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@agent/recognition/binarymatch.py` around lines 398 - 403, The example comment contains an outdated ROI for CheckPanel_A (still [640, 616, 17, 16]); update the example to the new value [723, 616, 18, 18] (or explicitly mark it as a placeholder) so the inline documentation matches the current detector behavior; specifically edit the commented example for "CheckPanel_A" in binarymatch.py to set CheckPanel_A.roi to [723, 616, 18, 18] and, if helpful, add a short note that RedDotDetector now crops the ROI and uses the new internal blob/confidence logic to avoid confusion.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@agent/recognition/binarymatch.py`:
- Around line 633-650: The code currently conflates reco is None (recognition
not started) with reco.hit == False (recognition ran but missed); update the
logic around Context.run_recognition/preset handling so you first check if reco
is None and return a distinct AnalyzeResult indicating the recognition did not
start (include context like preset_node and any available status), and only if
reco is non-None then branch on reco.hit; when reco.hit is False, return
CustomRecognition.AnalyzeResult(box=None, detail=...) but propagate the preset
node's raw_detail (e.g., reco.raw_detail) into the returned detail along with
result:"miss" and existing roi/hint fields so upstream can see stage/hint/stat
from the preset node rather than masking configuration errors as ordinary
misses.
- Around line 553-555: The module currently parses RDD_DUMP_INTERVAL at import
time into _DUMP_MIN_INTERVAL which can raise ValueError and abort import
(preventing RedDotDetector registration); change this to parse/validate with a
safe fallback at runtime (e.g., in the RedDotDetector initializer or before
first use) or wrap the environment parsing in a try/except that falls back to a
sane default and logs the parse error; ensure the symbol _DUMP_MIN_INTERVAL is
assigned only after validation so imports never raise, and reference
RDD_DUMP_INTERVAL, _DUMP_MIN_INTERVAL, and RedDotDetector (and analyze() if
relevant) when updating the code paths.
---
Outside diff comments:
In `@assets/resource/base/pipeline/Test.json`:
- Around line 668-714: The Agt_RedDot_Preset still uses removed thresholds
inner_v_min/inner_s_max while RedDot_Preset has been migrated to the v2 style
(min_confidence, gap_ratio, red_area, hsv_ranges, debug); update the
Agt_RedDot_Preset definition to the same v2 parameter set (replace
inner_v_min/inner_s_max with min_confidence and include gap_ratio, red_area,
hsv_ranges, debug as in RedDot_Preset) and then verify/adjust every node that
references {"preset":"Agt_RedDot_Preset"} (e.g., nodes similar to CheckPanel_A)
to either point to the updated Agt_RedDot_Preset or to RedDot_Preset so all
red-dot detections use the unified min_confidence-based thresholds.
---
Nitpick comments:
In `@agent/recognition/binarymatch.py`:
- Around line 398-403: The example comment contains an outdated ROI for
CheckPanel_A (still [640, 616, 17, 16]); update the example to the new value
[723, 616, 18, 18] (or explicitly mark it as a placeholder) so the inline
documentation matches the current detector behavior; specifically edit the
commented example for "CheckPanel_A" in binarymatch.py to set CheckPanel_A.roi
to [723, 616, 18, 18] and, if helpful, add a short note that RedDotDetector now
crops the ROI and uses the new internal blob/confidence logic to avoid
confusion.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 3929bcfa-538f-4d64-8f5a-0d607d67042c
📒 Files selected for processing (2)
agent/recognition/binarymatch.pyassets/resource/base/pipeline/Test.json
Contributor
🤖 DeepSeek 自动评审报告模型： 概览PR #299 将红点检测器（ 阻塞性问题（必须修改）1. 未命中时返回
|
具体在pr中标注.
Owner
Author
否决,与事实不符. |
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@agent/recognition/binarymatch.py`:
- Around line 650-665: The preset-mode AnalyzeResult detail is inconsistent: the
hit branch (CustomRecognition.AnalyzeResult in this file around the preset
handling for variable preset_node) only returns {"result","preset"} while the
miss branch embeds the preset node's stage/stat/conf/parts under "preset_detail"
using reco.raw_detail; unify both branches to the same detail schema promised by
the file-level docs by extracting the preset node's stage/stat/conf/parts (from
getattr(reco, "raw_detail", None) or reco.raw_detail) into the same top-level
detail fields (e.g., stage, stat, conf, parts or a single "preset" object) for
both hit and miss, preserve existing keys like
"result","mode","preset","roi","hint", and ensure the hit branch includes the
same preset-related fields so downstream code (MaaLogs/tuning scripts) can rely
on one schema.
- Around line 554-557: The environment parsing for _DUMP_MIN_INTERVAL should
treat NaN, Infinity, and non-positive values as invalid; when reading
RDD_DUMP_INTERVAL (currently via os.environ.get in the module scope), parse with
float inside the existing try/except, then validate with math.isfinite(value)
and value > 0 and fallback to 2.0 if validation fails; keep the existing
exception handling (TypeError, ValueError) and ensure the final assigned symbol
is _DUMP_MIN_INTERVAL with the safe default.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 5a72b615-c5cd-483a-ae86-78d5db084686
📒 Files selected for processing (1)
agent/recognition/binarymatch.py
Comment on lines
+554
to
+557
| try: | ||
| _DUMP_MIN_INTERVAL = float(os.environ.get("RDD_DUMP_INTERVAL", "2.0")) | ||
| except (TypeError, ValueError): | ||
| _DUMP_MIN_INTERVAL = 2.0 # 环境变量非法时兜底，避免 import 阶段崩溃中断 Agent 启动 |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
python - <<'PY'
import math
for raw in ("2.0", "nan", "inf", "-1", "oops"):
try:
value = float(raw)
print(raw, "->", value, "isfinite=", math.isfinite(value), "negative=", value < 0)
except Exception as exc:
print(raw, "->", type(exc).__name__)
PYRepository: sunyink/MFABD2
Length of output: 246
[minor] 让 RDD_DUMP_INTERVAL 对 nan/inf/负数 也视为非法并兜底（避免节流失效）
当前 float() 对 RDD_DUMP_INTERVAL=nan/inf 会解析成功而不会触发 try/except（math.isfinite 为 False），后续节流比较可能失效；同时负数也会通过解析并反向影响节流语义。
🩹 建议补丁
+import math
import json
import re
import time
import os
import traceback
@@
try:
- _DUMP_MIN_INTERVAL = float(os.environ.get("RDD_DUMP_INTERVAL", "2.0"))
+ _dump_min_interval = float(os.environ.get("RDD_DUMP_INTERVAL", "2.0"))
+ if not math.isfinite(_dump_min_interval) or _dump_min_interval < 0:
+ raise ValueError
+ _DUMP_MIN_INTERVAL = _dump_min_interval
except (TypeError, ValueError):
_DUMP_MIN_INTERVAL = 2.0 # 环境变量非法时兜底，避免 import 阶段崩溃中断 Agent 启动📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| try: | |
| _DUMP_MIN_INTERVAL = float(os.environ.get("RDD_DUMP_INTERVAL", "2.0")) | |
| except (TypeError, ValueError): | |
| _DUMP_MIN_INTERVAL = 2.0 # 环境变量非法时兜底，避免 import 阶段崩溃中断 Agent 启动 | |
| try: | |
| _dump_min_interval = float(os.environ.get("RDD_DUMP_INTERVAL", "2.0")) | |
| if not math.isfinite(_dump_min_interval) or _dump_min_interval < 0: | |
| raise ValueError | |
| _DUMP_MIN_INTERVAL = _dump_min_interval | |
| except (TypeError, ValueError): | |
| _DUMP_MIN_INTERVAL = 2.0 # 环境变量非法时兜底，避免 import 阶段崩溃中断 Agent 启动 |
🧰 Tools
🪛 Ruff (0.15.15)
[warning] 557-557: Comment contains ambiguous ， (FULLWIDTH COMMA). Did you mean , (COMMA)?
(RUF003)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@agent/recognition/binarymatch.py` around lines 554 - 557, The environment
parsing for _DUMP_MIN_INTERVAL should treat NaN, Infinity, and non-positive
values as invalid; when reading RDD_DUMP_INTERVAL (currently via os.environ.get
in the module scope), parse with float inside the existing try/except, then
validate with math.isfinite(value) and value > 0 and fallback to 2.0 if
validation fails; keep the existing exception handling (TypeError, ValueError)
and ensure the final assigned symbol is _DUMP_MIN_INTERVAL with the safe
default.
Comment on lines
+650
to
+665
| if reco.hit: | ||
| bx, by, bw, bh = reco.box | ||
| adjusted = (bx + rx, by + ry, bw, bh) | ||
| print(f"[RedDotDetector] [preset:{preset_node}] 命中 → {adjusted}") | ||
| return CustomRecognition.AnalyzeResult(box=adjusted, detail={"preset": preset_node}) | ||
| return None | ||
| mfaalog.info(f"[RedDotDetector] [preset:{preset_node}] hit -> {adjusted}") | ||
| return CustomRecognition.AnalyzeResult( | ||
| box=adjusted, detail={"result": "hit", "preset": preset_node}) | ||
|
|
||
| # 真未命中：阶段原因已由预设节点(独立模式)记进嵌套识别记录；这里附带透传其 raw_detail | ||
| raw = getattr(reco, "raw_detail", None) | ||
| mfaalog.warning(f"[RedDotDetector] miss@preset | {argv.node_name} via {preset_node}") | ||
| return CustomRecognition.AnalyzeResult(box=None, detail={ | ||
| "result": "miss", "mode": "preset", "preset": preset_node, | ||
| "roi": [rx, ry, rw, rh], | ||
| "preset_detail": raw, | ||
| "hint": "阶段原因见预设节点(独立模式)的 detail；失败截图见 debug/RedDotDetector/ 下以本节点名命名的 rdd_* 文件", | ||
| }) |
Contributor
There was a problem hiding this comment.
预设模式的 detail 还没有真正统一。
Line 654 只返回了 {"result", "preset"}，Line 663 又把预设节点的 stage/stat/conf/parts 包进了 preset_detail。这样预设模式命中/未命中的 detail 结构都和本文件上方文档承诺的统一 detail.* schema 不一致，接入 MaaLogs 或调参脚本时还得额外分支判断，等于把这次 PR 想统一的 preset/standalone 契约又拆开了。
♻️ 建议补丁
if reco.hit:
+ raw = getattr(reco, "raw_detail", None)
bx, by, bw, bh = reco.box
adjusted = (bx + rx, by + ry, bw, bh)
+ detail = raw.copy() if isinstance(raw, dict) else (
+ {"raw_detail": raw} if raw is not None else {}
+ )
+ detail.update({
+ "result": "hit",
+ "mode": "preset",
+ "preset": preset_node,
+ "roi": [rx, ry, rw, rh],
+ "box": list(adjusted),
+ })
mfaalog.info(f"[RedDotDetector] [preset:{preset_node}] hit -> {adjusted}")
- return CustomRecognition.AnalyzeResult(
- box=adjusted, detail={"result": "hit", "preset": preset_node})
+ return CustomRecognition.AnalyzeResult(box=adjusted, detail=detail)
# 真未命中：阶段原因已由预设节点(独立模式)记进嵌套识别记录；这里附带透传其 raw_detail
raw = getattr(reco, "raw_detail", None)
+ detail = raw.copy() if isinstance(raw, dict) else (
+ {"raw_detail": raw} if raw is not None else {}
+ )
+ detail.update({
+ "result": "miss",
+ "mode": "preset",
+ "preset": preset_node,
+ "roi": [rx, ry, rw, rh],
+ "hint": "阶段原因见预设节点(独立模式)的 detail；失败截图见 debug/RedDotDetector/ 下以本节点名命名的 rdd_* 文件",
+ })
mfaalog.warning(f"[RedDotDetector] miss@preset | {argv.node_name} via {preset_node}")
- return CustomRecognition.AnalyzeResult(box=None, detail={
- "result": "miss", "mode": "preset", "preset": preset_node,
- "roi": [rx, ry, rw, rh],
- "preset_detail": raw,
- "hint": "阶段原因见预设节点(独立模式)的 detail；失败截图见 debug/RedDotDetector/ 下以本节点名命名的 rdd_* 文件",
- })
+ return CustomRecognition.AnalyzeResult(box=None, detail=detail)🧰 Tools
🪛 Ruff (0.15.15)
[warning] 657-657: Comment contains ambiguous ： (FULLWIDTH COLON). Did you mean : (COLON)?
(RUF003)
[warning] 657-657: Comment contains ambiguous ； (FULLWIDTH SEMICOLON). Did you mean ; (SEMICOLON)?
(RUF003)
[warning] 664-664: String contains ambiguous ； (FULLWIDTH SEMICOLON). Did you mean ; (SEMICOLON)?
(RUF001)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@agent/recognition/binarymatch.py` around lines 650 - 665, The preset-mode
AnalyzeResult detail is inconsistent: the hit branch
(CustomRecognition.AnalyzeResult in this file around the preset handling for
variable preset_node) only returns {"result","preset"} while the miss branch
embeds the preset node's stage/stat/conf/parts under "preset_detail" using
reco.raw_detail; unify both branches to the same detail schema promised by the
file-level docs by extracting the preset node's stage/stat/conf/parts (from
getattr(reco, "raw_detail", None) or reco.raw_detail) into the same top-level
detail fields (e.g., stage, stat, conf, parts or a single "preset" object) for
both hit and miss, preserve existing keys like
"result","mode","preset","roi","hint", and ensure the hit branch includes the
same preset-related fields so downstream code (MaaLogs/tuning scripts) can rely
on one schema.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary by Sourcery
优化红点（通知）检测器，引入基于置信度的评分模型，提供更丰富的诊断信息和常驻的调试产物，并改进预设/独立模式下的行为以及错误报告。
Enhancements:
Original summary in English
Summary by Sourcery
Refine the red-dot (notification) detector to use a confidence-based scoring model with richer diagnostics and always-on debug artifacts, and improve preset/standalone behavior and error reporting.
Enhancements:
Summary by CodeRabbit
发布说明
新功能
Bug 修复
Chores