@pls-compile-owo/fisheye 0.8.0 → 0.8.2

package/CHANGELOG.md

@@ -2,6 +2,20 @@
 
 本项目遵循 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.1.0/) 规范与 [语义化版本](https://semver.org/lang/zh-CN/)。
 
+## [0.8.2] - 2026-11-14
+
+### Fixed
+
+- **广告段检测在稀疏弹幕视频上误报**：弹幕密度低于 0.2 条/秒时密度统计不可靠（艺术短片《利马》127 条/856s 被误标 3 段「低互动段」），改为直接跳过检测（宁漏勿错）。
+- golden fixture 补充人工真值标注（利马三段确认非广告）。
+
+## [0.8.1] - 2026-11-14
+
+### Added
+
+- **golden case 验收集**（`test/golden/`）：4 条真实 B站 视频（含字幕/无字幕/高密度弹幕/利马定格片）的字幕与弹幕样本落盘为 fixture，离线验证字幕质量判定、广告段检测、取帧计划的输出与真实回放一致。
+- 字幕质量判定提取为纯函数 `assessSubtitleQuality`（bili.mjs），供单测与 golden 用。
+
 ## [0.8.0] - 2026-11-14
 
 ### Changed

package/README.md

@@ -405,7 +405,7 @@ B 站的弹幕密度本身就是内容信号。实测一条 40 分钟视频：
 ## 开发
 
 ```bash
-node --test test/                # 单元测试（全部模块：46 个用例）
+node --test test/                # 单元测试 + golden 验收集（真实视频样本，离线可跑）
 node test-danmaku-filter.js      # 弹幕过滤规则测试（51 条用例）
 node fisheye.mjs selftest --deep # 含抽帧/字幕带实测的自检
 ```

package/lib/bili.mjs

@@ -76,6 +76,25 @@ export function pickSubtitleTracks(subs, preferZh = true) {
   return { main: foreign, translate: zh }
 }
 
+/**
+ * 字幕质量判定（纯函数，供单测/golden case 用）。
+ * 中文 AI 字幕在纯音乐/音效视频上会输出大量 ♪ 噪声或重复行——质量太差不能当台词。
+ * @param {string[]} contents 每行字幕文本（含 ♪ 符号也行）
+ * @returns {{ quality: 'ok'|'noisy', avgEff: number, uniqueRatio: number }}
+ */
+export function assessSubtitleQuality(contents) {
+  const body = Array.isArray(contents) ? contents.filter((c) => typeof c === 'string' && c.trim()) : []
+  if (!body.length) return { quality: 'noisy', avgEff: 0, uniqueRatio: 0 }
+  const eff = body.map((c) => c.replace(/[♪\s]/g, '').length)
+  const avgEff = eff.reduce((a, b) => a + b, 0) / body.length
+  const uniqueRatio = new Set(body.map((c) => c.replace(/[♪\s]/g, ''))).size / body.length
+  return {
+    quality: avgEff < 2 || uniqueRatio < 0.4 ? 'noisy' : 'ok',
+    avgEff: Number(avgEff.toFixed(1)),
+    uniqueRatio: Number(uniqueRatio.toFixed(2)),
+  }
+}
+
 export async function getOfficialSubs(bvid, cid, range = null) {
   try {
     const nav = await (await fetchWithRetry('https://api.bilibili.com/x/web-interface/nav',

package/lib/danmaku.mjs

@@ -6,6 +6,10 @@ export const AD_SEGMENT_SIGNAL = /金主|恰饭|商单|硬广|软广|赞助|推
 
 export function detectAdSegments(danmaku, duration, bucketSec = 30, stepSec = 10) {
   if (!danmaku?.length || duration < 120) return []
+  // 弹幕太稀疏（如艺术短片 127 条 / 856s ≈ 0.15 条/s）时密度统计不可靠：
+  // 中位数本身极低，「低于中位数 45%」的段落只会是「观众不爱发弹幕的正常段落」。
+  // 低于 0.2 条/秒直接不判广告（宁漏勿错，低互动段标签也会误导）。
+  if (danmaku.length / duration < 0.2) return []
   const dens = []
   for (let s = 0; s + bucketSec <= duration; s += stepSec) {
     let n = 0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pls-compile-owo/fisheye",
-  "version": "0.8.0",
+  "version": "0.8.2",
   "description": "Token-budget-aware video understanding for AI agents: official subtitles -> burned-in subtitle band -> danmaku highlights -> adaptive keyframe grids -> targeted refine.",
   "type": "module",
   "bin": {

package/test/golden/fixtures/BV1Cc411Q78T.json

@@ -0,0 +1,27 @@
+{
+  "bvid": "BV1Cc411Q78T",
+  "title": "起床，老婆，亲亲，三连【返场真喵真事】",
+  "duration": 30,
+  "hasOfficialSubs": true,
+  "expectedQuality": "ok",
+  "subtitleLines": [
+    "♪ 我想睡到自然醒 ♪",
+    "♪ 换到今天星期几 ♪",
+    "♪ 忘掉所有烦恼和无聊的问题 ♪",
+    "♪ 写着那首的厨艺 ♪",
+    "♪ 想你 ♪",
+    "♪ 忙碌了自己在追半个地燃烧我的卡路里 ♪"
+  ],
+  "expectedAdSegments": 0,
+  "danmaku": [
+    {
+      "t": 0,
+      "text": "猫界周棋洛doge"
+    },
+    {
+      "t": 3.431,
+      "text": "死亡音乐"
+    }
+  ],
+  "danmakuCount": 2
+}

package/test/golden/fixtures/BV1kr4y1x7DU.json

@@ -0,0 +1,521 @@
+{
+  "bvid": "BV1kr4y1x7DU",
+  "title": "屡获大奖！唯美催泪定格动画《利马》",
+  "duration": 856,
+  "hasOfficialSubs": false,
+  "expectedQuality": null,
+  "subtitleLines": [],
+  "expectedAdSegments": 0,
+  "danmaku": [
+    {
+      "t": 79.295,
+      "text": "我独自一人"
+    },
+    {
+      "t": 279.727,
+      "text": "你在吗"
+    },
+    {
+      "t": 367.006,
+      "text": "应该是父亲吧…"
+    },
+    {
+      "t": 811.499,
+      "text": "无须找寻，它自会来"
+    },
+    {
+      "t": 95.985,
+      "text": "我在陪你看"
+    },
+    {
+      "t": 420.144,
+      "text": "hello 10"
+    },
+    {
+      "t": 492.372,
+      "text": "这鸟有点意思"
+    },
+    {
+      "t": 780.839,
+      "text": "有暴风雨他去不了山顶看。于是雨停了。夜幕印在的水洼上或水塘。"
+    },
+    {
+      "t": 238.537,
+      "text": "他变成了星星。"
+    },
+    {
+      "t": 217.427,
+      "text": "果然。"
+    },
+    {
+      "t": 157.318,
+      "text": "呃呃。"
+    },
+    {
+      "t": 464.697,
+      "text": "感觉是缅怀父亲"
+    },
+    {
+      "t": 278.88,
+      "text": "对的"
+    },
+    {
+      "t": 103.536,
+      "text": "有人在嘛"
+    },
+    {
+      "t": 68.412,
+      "text": "2月6日晚在嘛"
+    },
+    {
+      "t": 48.639,
+      "text": "嗨"
+    },
+    {
+      "t": 780.291,
+      "text": "萤火虫又来了"
+    },
+    {
+      "t": 642.594,
+      "text": "好像我老家的房子啊，这种感觉很熟悉"
+    },
+    {
+      "t": 240.207,
+      "text": "是思亲吧"
+    },
+    {
+      "t": 240.207,
+      "text": "前面的小孩长大了，来到多年前和父亲晚上出来的那个地方，放了当时在父亲手中划过的萤火虫"
+    },
+    {
+      "t": 175.257,
+      "text": "恐怖来源于自己的内心和想象吧"
+    },
+    {
+      "t": 55.832,
+      "text": "我爱刘有生"
+    },
+    {
+      "t": 521.146,
+      "text": "木雕定格吧"
+    },
+    {
+      "t": 273.184,
+      "text": "这是定格吧"
+    },
+    {
+      "t": 77.34,
+      "text": "你哥"
+    },
+    {
+      "t": 103.44,
+      "text": "那样我就不孤独了"
+    },
+    {
+      "t": 93.595,
+      "text": "睡不着"
+    },
+    {
+      "t": 76.319,
+      "text": "是我呀！"
+    },
+    {
+      "t": 191.003,
+      "text": "角色形象和打光确实有点阴森"
+    },
+    {
+      "t": 84.455,
+      "text": "有人吗？"
+    },
+    {
+      "t": 94.372,
+      "text": "我也是。"
+    },
+    {
+      "t": 247.781,
+      "text": "好漂亮"
+    },
+    {
+      "t": 288.178,
+      "text": "我在"
+    },
+    {
+      "t": 243.262,
+      "text": "有点点忧伤"
+    },
+    {
+      "t": 101.566,
+      "text": "还有我"
+    },
+    {
+      "t": 47.764,
+      "text": "。"
+    },
+    {
+      "t": 565.465,
+      "text": "子欲养而亲不待，孤独的人渴望幸福"
+    },
+    {
+      "t": 305.283,
+      "text": "实在有点像蚯蚓的头"
+    },
+    {
+      "t": 173.543,
+      "text": "这建模建的有点奇怪"
+    },
+    {
+      "t": 664.779,
+      "text": "回忆 过去 现实"
+    },
+    {
+      "t": 159.025,
+      "text": "为什么我感觉好恐怖。。。"
+    },
+    {
+      "t": 570.58,
+      "text": "有一点点淡淡的压抑"
+    },
+    {
+      "t": 393.624,
+      "text": "什么意思呢"
+    },
+    {
+      "t": 368.993,
+      "text": "他爸好吧"
+    },
+    {
+      "t": 362.886,
+      "text": "？"
+    },
+    {
+      "t": 64.459,
+      "text": "在这纳"
+    },
+    {
+      "t": 816.318,
+      "text": "明月有情应照我"
+    },
+    {
+      "t": 122.6,
+      "text": "他在干吗？？"
+    },
+    {
+      "t": 69.042,
+      "text": "有"
+    },
+    {
+      "t": 56.9,
+      "text": "有人吗"
+    },
+    {
+      "t": 174.217,
+      "text": "我还是更喜欢那个导盲犬的动画"
+    },
+    {
+      "t": 80.316,
+      "text": "是我"
+    },
+    {
+      "t": 108.073,
+      "text": "mc?"
+    },
+    {
+      "t": 6.398,
+      "text": "尼玛"
+    },
+    {
+      "t": 55.799,
+      "text": "前面的我在这里"
+    },
+    {
+      "t": 81.559,
+      "text": "自己人"
+    },
+    {
+      "t": 809.262,
+      "text": "空巢老人了"
+    },
+    {
+      "t": 718.602,
+      "text": "他没有这种记忆"
+    },
+    {
+      "t": 456.782,
+      "text": "日月如梭"
+    },
+    {
+      "t": 105.762,
+      "text": "他美好的记忆几乎没有吧"
+    },
+    {
+      "t": 835.988,
+      "text": "谁能想到这是用♂做的动漫"
+    },
+    {
+      "t": 792.974,
+      "text": "扭曲的窗子"
+    },
+    {
+      "t": 744.434,
+      "text": "没做什么大错，却要很多磨难的人有很多"
+    },
+    {
+      "t": 681.814,
+      "text": "果然还是一个人呆着是最舒适的"
+    },
+    {
+      "t": 653.214,
+      "text": "阿凡提"
+    },
+    {
+      "t": 522.294,
+      "text": "阿凡提风格"
+    },
+    {
+      "t": 473.894,
+      "text": "书包？"
+    },
+    {
+      "t": 365.242,
+      "text": "不是伴侣，是他爸"
+    },
+    {
+      "t": 284.302,
+      "text": "根据我上早读，同桌捣乱，多嘈杂也能静下来的经验，可能会有些干扰，不过应该也差不多"
+    },
+    {
+      "t": 572.83,
+      "text": "小鸟陪着他呢"
+    },
+    {
+      "t": 507.97,
+      "text": "变成鸟飞走了"
+    },
+    {
+      "t": 70.215,
+      "text": "黑的只能看到我自己"
+    },
+    {
+      "t": 374.03,
+      "text": "老爹的老爹"
+    },
+    {
+      "t": 792.679,
+      "text": "一边看着外边的风景，一边警示着外边"
+    },
+    {
+      "t": 696.819,
+      "text": "大鼻子啊"
+    },
+    {
+      "t": 649.559,
+      "text": "阿梵达风格既视感"
+    },
+    {
+      "t": 2.153,
+      "text": "利马＝nm"
+    },
+    {
+      "t": 75.899,
+      "text": "这里"
+    },
+    {
+      "t": 73.379,
+      "text": "我"
+    },
+    {
+      "t": 74.838,
+      "text": "是谁"
+    },
+    {
+      "t": 73.754,
+      "text": "我。。。"
+    },
+    {
+      "t": 51.781,
+      "text": "前面wei的，我在这儿"
+    },
+    {
+      "t": 84.434,
+      "text": "你不是一个人"
+    },
+    {
+      "t": 7.923,
+      "text": "nm"
+    },
+    {
+      "t": 120.659,
+      "text": "讲大青蛙的吗？"
+    },
+    {
+      "t": 249.779,
+      "text": "场景很美"
+    },
+    {
+      "t": 77.429,
+      "text": "我"
+    },
+    {
+      "t": 240.26,
+      "text": "他刚刚在怀念小时候跟爸爸"
+    },
+    {
+      "t": 560.322,
+      "text": "好孤单"
+    },
+    {
+      "t": 245.322,
+      "text": "耐心看完"
+    },
+    {
+      "t": 209.042,
+      "text": "眼睛"
+    },
+    {
+      "t": 251.459,
+      "text": "就是想他爹了吧"
+    },
+    {
+      "t": 81.485,
+      "text": "我"
+    },
+    {
+      "t": 76.739,
+      "text": "鬼和你一起看"
+    },
+    {
+      "t": 818.682,
+      "text": "蜡烛熄了，他也去了另一个世界"
+    },
+    {
+      "t": 79.202,
+      "text": "我"
+    },
+    {
+      "t": 100.059,
+      "text": "还有我"
+    },
+    {
+      "t": 235.796,
+      "text": "所以有没人来告诉我它讲得啥"
+    },
+    {
+      "t": 638.061,
+      "text": "我瞎了.非要搞得阴暗神秘"
+    },
+    {
+      "t": 77.078,
+      "text": "我"
+    },
+    {
+      "t": 82.997,
+      "text": "还有我"
+    },
+    {
+      "t": 74.387,
+      "text": "我"
+    },
+    {
+      "t": 74.019,
+      "text": "我"
+    },
+    {
+      "t": 43.339,
+      "text": "wei'"
+    },
+    {
+      "t": 675.739,
+      "text": "无助"
+    },
+    {
+      "t": 646.316,
+      "text": "孤独 寂寞 冷"
+    },
+    {
+      "t": 567.623,
+      "text": "他也老了"
+    },
+    {
+      "t": 359.29,
+      "text": "逝去的伴侣"
+    },
+    {
+      "t": 828.642,
+      "text": "4"
+    },
+    {
+      "t": 77.511,
+      "text": "不止一个"
+    },
+    {
+      "t": 450.247,
+      "text": "距离以一颗观月的心"
+    },
+    {
+      "t": 69.45,
+      "text": "是谁和我一起看呢。"
+    },
+    {
+      "t": 423.437,
+      "text": "你好"
+    },
+    {
+      "t": 438.201,
+      "text": "思念是一条长河"
+    },
+    {
+      "t": 514.953,
+      "text": "这不是木雕动画吗"
+    },
+    {
+      "t": 423.853,
+      "text": "迎客松~"
+    },
+    {
+      "t": 263.437,
+      "text": "pump stars"
+    },
+    {
+      "t": 802.209,
+      "text": "哇"
+    },
+    {
+      "t": 597.169,
+      "text": "小心啊"
+    },
+    {
+      "t": 101.369,
+      "text": "da打开就看不到星空啦"
+    },
+    {
+      "t": 495.651,
+      "text": "哈哈哈这只鸟还会眨眼睛"
+    },
+    {
+      "t": 189.205,
+      "text": "白蚁？"
+    },
+    {
+      "t": 698.899,
+      "text": "哈喽5"
+    },
+    {
+      "t": 409.939,
+      "text": "哈喽，10"
+    },
+    {
+      "t": 344.56,
+      "text": "去月球"
+    },
+    {
+      "t": 166.497,
+      "text": "塌鼻格魯"
+    },
+    {
+      "t": 98.354,
+      "text": "好想把灯打开"
+    }
+  ],
+  "danmakuCount": 127,
+  "groundTruthNote": "2026-11-14 人工确认：04:50-05:50 / 08:50-10:40 / 11:40-13:00 三段均非广告（艺术短片，无赞助），修复稀疏弹幕误报后应检测为 0 段"
+}