Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code:   DatasetGenerationError
Exception:    ValueError
Message:      Expected object or value
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1827, in _prepare_split_single
                  for key, table in generator:
                                    ^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
                  for item in generator(*args, **kwargs):
                              ~~~~~~~~~^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 281, in _generate_tables
                  examples = [ujson_loads(line) for line in batch.splitlines()]
                              ~~~~~~~~~~~^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
                  return pd.io.json.ujson_loads(*args, **kwargs)
                         ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
              ValueError: Expected object or value
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1880, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

messages
list
uid
string
task
string
dataset
string
duration_s
float64
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/STAR/W1SRN.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video reasoning syst...
c487da2ef9c526bb
verifier
Video-R1/STAR
32.399
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/STAR/2074D.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video reasoning syst...
89a738d3bff6d593
verifier
Video-R1/STAR
22.719
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/PerceptionTest/video_5066.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video...
8f2c93377e184e67
verifier
Video-R1/PerceptionTest
26.997
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_R8kLRs4QfEQ.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
3832a345a754c9b6
verifier
LLaVA-Video/30_60_s_youtube_v0_1
37.133
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_SbPU2e_bKjQ.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
ec4f4e362d3b9c96
verifier
LLaVA-Video/30_60_s_youtube_v0_1
58.959
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_Jve2ts4JasI.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
d1f7aa5705f36d7a
verifier
LLaVA-Video/30_60_s_youtube_v0_1
49.6
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/CLEVRER/train_videos/video_05782.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angl...
6b411fe34894ae8f
verifier
Video-R1/CLEVRER
5.12
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/CLEVRER/train_videos/video_08934.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angl...
47ab04bd84539fd4
verifier
Video-R1/CLEVRER
5.12
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_eibPOw8G1Lg.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
d2a6e564528587f4
verifier
LLaVA-Video/30_60_s_youtube_v0_1
51.567
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_AhtYiOEDF_U.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
98391d2b048c95f4
verifier
LLaVA-Video/30_60_s_youtube_v0_1
58.367
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_pFfWaV4Xn7U.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
54a9ca8dc0a4ac8b
verifier
LLaVA-Video/30_60_s_youtube_v0_1
48.88
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/PerceptionTest/video_5096.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video...
785771217f8accad
verifier
Video-R1/PerceptionTest
30.943
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/PerceptionTest/video_2352.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video...
476a889de7236408
verifier
Video-R1/PerceptionTest
22.034
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_CRGk4JCZRCQ.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
8967ae51a60a52b8
verifier
LLaVA-Video/30_60_s_youtube_v0_1
42.867
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_eNx6puTETG8.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
3b8b87264d12a5f7
verifier
LLaVA-Video/30_60_s_youtube_v0_1
36.667
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_6JZ4dVSbxrs.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
f5adce3d1e74dd90
verifier
LLaVA-Video/30_60_s_youtube_v0_1
50.684
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_fEYQRb36Tqg.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
9b9f7673c0d39346
verifier
LLaVA-Video/30_60_s_youtube_v0_1
58.967
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_whmE1tsLqqE.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
c73fb9ee1d679cd6
verifier
LLaVA-Video/30_60_s_youtube_v0_1
44.6
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_Ct7IwPWI8N4.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
b1ec1d309701b435
verifier
LLaVA-Video/30_60_s_youtube_v0_1
34.567
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_AAIYPIcwyHI.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
1aae37413408b38c
verifier
LLaVA-Video/30_60_s_youtube_v0_1
30.533
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_S5dEoTL6Bu4.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
014301b4c57dd1f0
verifier
LLaVA-Video/30_60_s_youtube_v0_1
58.033
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_nErOoxGyQWA.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
f00edfb2c8b0fc32
verifier
LLaVA-Video/30_60_s_youtube_v0_1
44.067
[ { "role": "user", "content": [ { "type": "video", "video": "cgbench/videos/wL99SAaw3sw.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video reasoning sys...
06b3060eefe91288
verifier
CG-Bench
987.167
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_Lixi5PH8vSg.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
3af061c072cbdb1b
verifier
LLaVA-Video/30_60_s_youtube_v0_1
58.033
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/CLEVRER/train_videos/video_06977.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angl...
6d8428ee33916cbd
verifier
Video-R1/CLEVRER
5.12
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/PerceptionTest/video_3771.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video...
d5797cc2daeb2711
verifier
Video-R1/PerceptionTest
25.848
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_XeFbBxGtOiI.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
120f54270b6ae793
verifier
LLaVA-Video/30_60_s_youtube_v0_1
55.533
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_po-dVMdDPu0.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
54dca44030083e59
verifier
LLaVA-Video/30_60_s_youtube_v0_1
55.367
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_npZuayv7Bhg.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
cfa679b736cab146
verifier
LLaVA-Video/30_60_s_youtube_v0_1
58.933
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_TRS9u6Np8Pg.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
24af9f717a48b0e0
verifier
LLaVA-Video/30_60_s_youtube_v0_1
56.4
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/PerceptionTest/video_10702.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle vide...
3e72ec87ec164971
verifier
Video-R1/PerceptionTest
26.989
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_uWBgbZswZig.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
3d30329c1a254bca
verifier
LLaVA-Video/30_60_s_youtube_v0_1
43.667
[ { "role": "user", "content": [ { "type": "video", "video": "cgbench/videos/BV1vc411x7HD.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video reasoning sy...
6615823717d6906f
verifier
CG-Bench
2,164.662
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_P-6ucqlUm4I.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
f50a0e045ac3238f
verifier
LLaVA-Video/30_60_s_youtube_v0_1
48.9
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_-unC8xjf8NA.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
7c352aab175cee1a
verifier
LLaVA-Video/30_60_s_youtube_v0_1
38.172
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_4EbRtU4z-nU.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
b9b1135bc21017a1
verifier
LLaVA-Video/30_60_s_youtube_v0_1
38.405
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_4MHMABLio6Q.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
3794d464dfcc2e79
verifier
LLaVA-Video/30_60_s_youtube_v0_1
59.067
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_V3bQhK_lrzs.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
bafc6a539892a09b
verifier
LLaVA-Video/30_60_s_youtube_v0_1
59.333
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_IhFp0EOKQQE.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
adcb87a9d945de7e
verifier
LLaVA-Video/30_60_s_youtube_v0_1
58.058
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_R4siF4iIX6s.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
b75003fb583c058d
verifier
LLaVA-Video/30_60_s_youtube_v0_1
50.067
[ { "role": "user", "content": [ { "type": "video", "video": "llava_video/30_60_s_youtube_v0_1/liwei_youtube_videos/videos/youtube_video_2024/ytb_xajVCOOE6zs.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", ...
e2fd7a085e695580
verifier
LLaVA-Video/30_60_s_youtube_v0_1
34.833
[ { "role": "user", "content": [ { "type": "video", "video": "videor1/videos/NeXT-QA/NextQA/NExTVideo/1111/2840884667.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a mu...
f6ac96c20b1bd6d3
verifier
Video-R1/NeXT-QA
70.704
[ { "role": "user", "content": [ { "type": "video", "video": "cgbench/videos/zPuLGkluoTY.mp4", "fps": 2, "max_frames": 256, "total_pixels": 25165824 }, { "type": "text", "text": "You are the verifier in a multi-angle video reasoning sys...
acdff5d615ab8397
verifier
CG-Bench
2,466.033
End of preview.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

RunningBench 人工标注 · 标注说明

本仓库是 RunningBench 人工标注的分发包。10 个包 annotation_bundles/bundle_01.tar … bundle_10.tar,共 1526 题,每包约 150 题, 按题源分层(fullvideo / excerpt / p01ma_gdrive / p01ma_hf / rbma273)。包内不含标准答案。

English version → README_EN.md


1. 你要做什么(操作流程)

任务:每题给出若干视频片段和 6–8 个选项,题目要求选恰好 n 项(多为 3 项)。你需要看视频后判断每一个选项是否被画面支持, 给出最终答案,并判断这道题本身是否清晰。不做盲答,直接看视频。

准备:下载你分到的那个 tar,解压。不需要安装任何软件,不需要联网。

步骤

  1. 双击 review.html,右上角填写你的姓名(导出文件会带上)。
  2. 每题四步:
    • ① 看:把该题的全部片段看完,边看边在纸上或文本里记一条时间线:转向(方向、大概角度)、路面变化、带数字/文字/颜色的物体、 行人车辆、显著建筑。时间不用精确,±3 秒即可,要的是先后顺序和大致位置。多片段题每个片段单独记一条, 注意 CLIP 标签顺序不代表时间顺序(CLIP_A 不一定比 CLIP_B 早)。
    • ② 逐选项核对:对每个选项点一个标记,不许凭整体印象:
      • ✓ 支持:画面里找得到支撑它的那几秒。
      • ✗ 排除:画面明确否定它。一处细节错即为错(左右反了、顺序反了、颜色不对)。画面覆盖了该地点/时刻但没有该物体,也算排除。
      • ? 无法判定:画面根本没拍到该选项说的地点/时刻。
    • ③ 最终答案:勾恰好 n 项,默认跟随 ✓。✓ 数量 ≠ n 时不要硬凑:填你最有把握的 n 项,判定选 ambiguous,备注写清是哪个选项导致的。
    • ④ 判定 + 备注:
      • clear:只有一组说得通的答案,且画面看得见。
      • ambiguous:存在第二组说得通的答案,或题干有歧义,或 ✓ 数量与 n 不符。
      • cant_tell:画面模糊、缺失,或没覆盖到关键时刻。
      • 备注写:哪个选项有问题、你看到了什么、大约第几秒。这是最有价值的反馈。
  3. 做完一题卡片变绿;页面顶部可筛选"未完成"。进度自动保存在本机浏览器,随时关闭、下次接着做。 换电脑:先点「导出 JSON」,在新电脑上用「导入进度」恢复。
  4. 全部完成后点右上「导出 JSON」,把下载到的 annotation_<姓名>_<日期>.json 回传。

工作量:单片段题 2–3 分钟,多片段比较题 4–6 分钟,一天约 80–100 题。每 90 分钟休息一次,疲劳会明显降低对近似干扰项的识别率。

片段命名:CLIP_1, CLIP_2… 是同一段录像按时间顺序切出的多段;CLIP_A…CLIP_J 是题干里点名的匿名片段,顺序与时间无关; VIDEO_A…VIDEO_R 是跨录像比较题里题干用到的匿名录像编号。


2. 怎么看、怎么判(校验手册全文)

本流程不做盲答,下文凡提到"盲答 / Step 1"的句子请忽略,其余全部适用。

0. 一条总原则

逐选项对账,不做整体印象判断。

错误做法:看完视频,凭整体感觉选出"对的三个"。 正确做法:8 个选项一个一个过,每个选项在视频里找到"支持它的画面"或 "否定它的画面",找不到就标记为"判不了"。

为什么:这批题的干扰项是故意造的 near-miss —— 拿画面里真实存在的东西, 改掉一两处细节(左右对调、顺序颠倒、颜色写错)。凭印象看,near-miss 全都"眼熟", 你会漏掉被改的那一两处。只有逐项对账能抓出来。


1. 两遍看法

第一遍 · 连续看完,记时间线

不看选项(或看过就先放下),从头到尾连续播放,手上记一条时间线:

0:05  出发,柏油路,路灯亮着
0:18  右转,进入鹅卵石路面
0:31  左侧出现编号 11 的建筑
0:44  对面骑车人(车前灯亮)驶过
0:58  经过自行车停放架
1:12  两根红白条纹护柱之间穿过
1:27  右侧混凝土长椅
1:40  红白施工围挡开始

记什么:转弯(方向+大概角度)、路面变化、有编号/文字/颜色的物体、 移动的人和车、明显的建筑。不用记全,记"可能被出题的东西"。

时间戳不用精确,±3 秒足够 —— 你要的是顺序和大概位置,不是秒表。

多片段的题(题干里有 CLIP_B、CLIP_F 这类标签):每个片段各记一条时间线, 标签写在开头。注意:标签顺序不代表时间顺序,CLIP_A 未必比 CLIP_B 早。

第二遍 · 逐选项定点核对

现在打开选项,一个一个过。每个选项:

  1. 拆出它声称的事实。比如选项 B: "穿过两根条纹护柱进入有混凝土长椅的路段,早于经过红白施工围挡" 拆成三件事:① 有条纹护柱 ② 护柱后有长椅 ③ 这些都在施工围挡之前
  2. 对照你的时间线。护柱 1:12、长椅 1:27、围挡 1:40 —— 三件事都对 → 支持
  3. 拿不准就拖回去看。用播放器拖到时间线上对应位置,只看那几秒。 这就是记时间线的意义:第二遍不用重看全片,只做定点核查。

每个选项在纸上标一个符号:✓ 支持 · ✗ 否定(写明哪一处错)· ? 判不了

全部 8 个过完再下结论。✓ 应该恰好 3 个 —— 多了少了都说明有问题 (要么标注错,要么题目歧义),这本身就是你要报告的发现。


2. 每类题看什么

题型 第一遍重点记 第二遍核对什么
event-order 事件顺序 每个显著事件的时间戳 选项里"X 早于 Y"逐对核对时间线
turn-pattern 转弯模式 每次转弯:方向、急缓、路口特征 选项说的方向/角度/次序有没有写反
landmark-order 地标顺序 地标出现时刻 + 在路的哪一侧 顺序 + 左右侧(最常被改的细节)
landmark-revisit 重复经过 同一物第二次出现的时刻和方向 是不是真的同一个物(对细节:颜色/文字/损伤)
environment-* 环境变化 路面材质/植被/建筑密度的切换点 变化方向对不对("从柏油到碎石"还是反过来)
spatial-consistency 空间关系 关键物在路的左/右/前方 去程右侧 = 回程左侧,别被折返骗了
same-place-different-recording 两个片段各自的地标清单 逐个地标比对:同一颗树、同一块牌子?
route-identity 是否同一条路 各片段的路面、转弯序列 转弯序列一致才算同路,单个地标相似不够

跨录像的题(对比两次不同录制)额外注意:

  • 光照可以完全不同(一次白天一次夜里),别因为"看起来不像"就判不同地点
  • 判"同一地点"要靠不变的东西:建筑形状、路的走向、固定设施
  • 判"不同"要靠结构差异:转弯序列不同、路面材质不同,而不是行人车辆这些偶然物

3. 判定规则(容易搞错的三种情况)

① 选项说的东西画面里根本没有

看情况,这是最容易判错的一类:

  • 片段完整覆盖了选项声称的位置/时段,那里没这东西 → ✗ 否定。 例:选项说"深灰色旅行车停在长椅旁",片段清楚拍到了长椅一带,没有任何车 → 否定。
  • 片段没拍到选项说的位置/时段 → ? 判不了。 例:选项讲"过桥之后…"而片段在过桥前就结束了 → 判不了。

区别在于:画面给没给你否定它的机会。

② 选项大部分对、一处错

→ ✗ 否定。 一处错就是错。near-miss 干扰项就长这样, "长椅在右侧"(实际在左侧)不能因为"确实有长椅"就放过。

③ 你看到的和标注答案数对不上

数出 4 个 ✓ 或只有 2 个 ✓ 时,不要硬凑成 3 个。如实选:

  • 你的三个最有把握的作为"最终答案"
  • 判定选 ambiguous(多于 3 个说得通)或 wrong_answer
  • 备注里写清是哪个选项出的问题 —— 这是最有价值的反馈

4. 完整示范(真题)

题型 event-order · 2 个片段(CLIP_C、CLIP_F,同一夜跑录像的两段)

Q: 以下哪些选项正确描述了路线上事件的时间顺序?选三个。

A. 在到达标有数字 11 的建筑之前,先与亮着车前灯的迎面骑车人擦肩而过 B. 穿过两根条纹护柱进入有混凝土长椅的路段,早于经过红白施工围挡 C. 经过红白施工围挡早于穿过两根条纹护柱 ← 注意:恰好是 B 的反命题 D. 在沿着经过施工围挡的直路之后,才在路右缘遇到混凝土长椅 E. 在鹅卵石路段经过 11 号建筑,早于到达室外自行车架 F. 迎面骑车人是在有施工围挡的直路柏油段遇到的,而非弯曲鹅卵石路段 G. 先经过 11 号建筑,之后才在鹅卵石路上遇到迎面骑车人 ← A 的反命题 H. 在鹅卵石路上经过自行车架,早于走上路灯下草坪相夹的直路

第一遍,两个片段各记时间线:

CLIP_C: 0:03 鹅卵石路 · 0:15 迎面骑车人(车灯) · 0:29 建筑"11" · 0:41 自行车架
CLIP_F: 0:05 条纹护柱×2 · 0:12 混凝土长椅(右) · 0:limit 草坪直路(路灯) · 0:47 施工围挡

还差一件事:CLIP_C 和 CLIP_F 谁先谁后? 标签不代表顺序。 线索:CLIP_H 段选项说"鹅卵石路早于草坪直路" —— 你得从画面推 (比如 CLIP_C 结束处的场景是否衔接 CLIP_F 开头,或看两段的疲劳程度/环境延续性), 推不出就依靠单段内部能判定的选项。

第二遍逐项:

A  骑车人(C 0:15) 早于建筑11(C 0:29)      → ✓ 同段内可判
B  护柱(F 0:05) 早于围挡(F 0:47)          → ✓ 同段内可判
C  与 B 相反                              → ✗
D  长椅(F 0:12) 在围挡(F 0:47) 之前,选项说在后 → ✗
E  建筑11(0:29) 早于车架(0:41)?选项方向对 → 等等,再核一遍…✓? 但标注答案没有 E!
   → 重看 0:29-0:41:车架其实 0:26 就在画面边缘出现过 → E 实际为 ✗
F  骑车人在鹅卵石段(C),选项说在柏油围挡段  → ✗
G  与 A 相反                              → ✗
H  车架(C 0:41,鹅卵石) 早于草坪直路(F)    → ✓ 需要跨段顺序,由画面衔接判定

✓ = A、B、H,恰好 3 个,与标注一致 → 判定 correct。

注意 E 那一步:差点被自己的时间线骗了 —— 第一遍漏记了 0:26 画面边缘的车架。 所以拿不准的选项一定要拖回去重看,时间线是索引,不是证据。


5. 常见错误清单

错误 后果
凭整体印象选三个,不逐项对账 near-miss 干扰项全漏
先看视频再补盲猜 盲猜数据作废,可猜率虚低
"画面里没有"一律判否定 把"判不了"误当"否定",高估标注质量
折返路段忘了左右互换 spatial 类题全判反
把 CLIP 标签顺序当时间顺序 event-order 题判反
凑答案数(明明看到 4 个对,硬选 3 个) 掩盖了题目歧义,这正是要报告的
跨录像题因光照不同判"不同地点" 昼夜对比题全错

6. 一天能校多少

熟练后单片段题约 2-3 分钟/道,多片段对比题 4-6 分钟/道。 一天 6 小时约 80-100 道。 抽样 120 道(推荐)约一天半。

疲劳会显著降低 near-miss 的辨识率 —— 连续校验不要超过 90 分钟,休息再来。


3. 回传与验收

  • 回传:每人一个导出的 JSON 文件,不需要改任何内容。
  • 验收线:标注一致率 > 90%(由回收方用标准答案合并计算)。一致率低不是你的问题,是题的问题,如实报告最重要。
Downloads last month
316