Datasets:

CedPei
/

TempCloze

source stringclasses 7 values	id stringlengths 9 28	gap_start float64 3.22 39.2	gap_end float64 6.55 64.4
care	v_00000203_5.mp4	5.6	10.4
care	v_00001358_22.mp4	5.57	9.25
care	v_00001379_10.mp4	7.31	11.4
care	v_00001383_2.mp4	4.83	7.83
care	v_00001387_0.mp4	7.29	14.2
care	v_00001859_1.mp4	4.64	7.79
care	v_00001948_5.mp4	3.6	6.97
care	v_00001979_12.mp4	3.83	8.81
care	v_00002024_0.mp4	8.72	14.98
care	v_00002077_10.mp4	6.54	10.02
care	v_00002077_26.mp4	6.26	10.64
care	v_00002177_0.mp4	9.76	18.38
care	v_00002188_4.mp4	14.93	21.36
care	v_00002240_4.mp4	4.31	8.11
care	v_00002271_1.mp4	4.68	10.62
care	v_00002283_1.mp4	4.69	9.42
care	v_00003197_0.mp4	5.52	8.05
care	v_00003574_0.mp4	4.67	8.94
care	v_00003675_0.mp4	5.55	9.28
care	v_00003905_0.mp4	6.22	10.96
care	v_00004005_0.mp4	7.1	17.5
care	v_00004014_0.mp4	10.27	23.28
care	v_00004018_3.mp4	6.49	11.44
care	v_00004040_0.mp4	11.28	22.95
care	v_00004040_1.mp4	15.8	29.03
care	v_00004043_0.mp4	14.36	31.37
care	v_00004629_0.mp4	7.03	12.65
care	v_00004635_0.mp4	8.49	18.11
care	v_00004656_5.mp4	8.61	18.79
care	v_00005151_0.mp4	6.2	9.47
care	v_00005181_0.mp4	4.86	9.38
care	v_00005268_0.mp4	5.02	8.41
care	v_00005279_0.mp4	6.92	10.24
care	v_00005377_0.mp4	5.41	9.05
care	v_00005709_0.mp4	3.73	6.84
care	v_00005890_3.mp4	5.42	12.16
care	v_00005910_8.mp4	5.08	10.78
care	v_00006264_1.mp4	4.75	9.89
care	v_00006334_2.mp4	9.08	20.77
care	v_00006464_0.mp4	6.38	11.3
care	v_00006509_0.mp4	7.07	14.94
care	v_00006606_0.mp4	4.47	8.46
care	v_00006670_0.mp4	3.22	6.94
care	v_00006695_0.mp4	7.72	15.68
care	v_00006705_0.mp4	10.47	15.91
care	v_00006711_0.mp4	5.75	11.94
care	v_00006714_0.mp4	8.2	16.38
care	v_00006963_0.mp4	4.05	8.24
care	v_00007041_10.mp4	4.78	9.72
care	v_00007120_5.mp4	9.84	16.27
care	v_00007126_5.mp4	9.14	17.93
care	v_00007217_0.mp4	10.21	18.26
care	v_00007235_2.mp4	6.12	11.21
care	v_00007535_0.mp4	6.94	11.34
care	v_00008644_0.mp4	18.33	27.24
care	v_00008647_0.mp4	5.75	8.68
care	v_00009012_0.mp4	6.48	10.23
care	v_00009404_0.mp4	4.66	10.39
care	v_00009485_0.mp4	5.23	10.77
care	v_00009537_0.mp4	5.89	12.83
care	v_00009586_0.mp4	4.23	9.21
care	v_00009591_0.mp4	3.61	8.81
care	v_00009689_0.mp4	6.17	10.36
care	v_00009690_0.mp4	6.1	9.49
care	v_00009721_0.mp4	7.58	10.55
care	v_00009731_0.mp4	4.84	8.33
care	v_00009795_0.mp4	7.62	14.09
care	v_00009911_0.mp4	4.38	11
care	v_00010075_0.mp4	4.82	8.1
care	v_00010607_0.mp4	6.25	11.51
care	v_00011221_1.mp4	8.46	15.71
care	v_00011273_0.mp4	6.81	14.36
care	v_00011376_0.mp4	6.07	9.78
care	v_00011378_0.mp4	10.01	18.29
care	v_00012658_2.mp4	7.48	10.58
care	v_00012676_1.mp4	5.06	9.55
care	v_00013082_0.mp4	6.26	10.53
care	v_00013169_0.mp4	6.93	14.06
care	v_00013363_0.mp4	4.2	7.07
care	v_00013624_0.mp4	7.18	10.72
care	v_00013697_3.mp4	4.6	8.33
care	v_00013757_0.mp4	7.02	11.54
care	v_00013853_0.mp4	4.95	10.37
care	v_00014469_3.mp4	4.13	8.19
care	v_00014883_2.mp4	5.56	8.82
care	v_00014911_0.mp4	6.69	9.84
care	v_00015005_0.mp4	3.82	6.59
care	v_00015029_0.mp4	5.03	7.77
care	v_00015039_1.mp4	6.43	10.09
care	v_00016057_5.mp4	6.54	9.54
care	v_00016063_6.mp4	4.05	7.88
care	v_00016220_0.mp4	6.57	11.81
care	v_00016880_0.mp4	6.78	17.15
care	v_00016895_0.mp4	9.85	18.21
dailyomni	0HzbpwB3xDk_video.mp4	11.42	22.43
dailyomni	0Mba3BS1oRM_video.mp4	10.63	18.89
dailyomni	0QYedLfOwcI_video.mp4	13.08	21.65
dailyomni	0izHOfrwPn4_video.mp4	12.62	21.16
dailyomni	0rz3h0ghprk_video.mp4	10.72	21.67
dailyomni	2__T6Q8rCCA_video.mp4	12.06	20.57

End of preview. Expand in Data Studio

TempCloze

Paper: TempCloze: Can Video-LLMs Identify the Missing Middle?

TempCloze is a video cloze benchmark for evaluating whether Video-LLMs can identify the missing middle of a video from its beginning and ending context.

This repository provides metadata for 1,521 videos from seven sources. Each row identifies one source video and the temporal boundaries of its missing segment. The benchmark code and evaluation instructions are available in the official GitHub repository.

Dataset structure

The test split contains the following fields:

Field	Type	Description
`source`	string	Short name of the source dataset
`id`	string	Original source video filename
`gap_start`	float	Missing segment start time in seconds
`gap_end`	float	Missing segment end time in seconds

The three video segments are determined directly from the source video and timestamps:

beginning: [0, gap_start)
ground truth: [gap_start, gap_end)
ending: [gap_end, video_end]

Example:

{"source":"care","id":"v_00000203_5.mp4","gap_start":5.6,"gap_end":10.4}

Dataset composition

Source	Videos
LVD-2M	515
EgoLife	437
MiraData	198
FAVOR-Bench	145
CaReBench	94
Video-TT	89
Daily-Omni	43
Total	1,521

Loading the metadata

from datasets import load_dataset

dataset = load_dataset("CedPei/TempCloze", split="test")
print(dataset[0])

Citation

@article{pei2026tempcloze,
  title   = {TempCloze: Can Video-LLMs Identify the Missing Middle?},
  author  = {Pei, Wenqi and Zhao, Henry Hengyuan and Liu, Yilai and Meng, Jiahao and Chen, Han and Wang, Ziyu and Du, Hongyang},
  journal = {arXiv preprint arXiv:2609.01515},
  year    = {2026}
}

Downloads last month: 94

Paper for CedPei/TempCloze

TempCloze: Can Video-LLMs Identify the Missing Middle?

Paper • 2609.01515 • Published 13 days ago • 29