Dataset Viewer
Auto-converted to Parquet Duplicate
video
video
5.47
29.7
video_id
stringlengths
28
38
width
int64
848
848
height
int64
480
480
fps
float64
24
30
fps_rational
stringclasses
6 values
num_frames
int64
137
890
duration_seconds
float64
5.47
29.7
size_bytes
int64
744k
9.13M
codec
stringclasses
1 value
-7wwfGJXEZg_frame169_fps25.0
848
480
25
25/1
137
5.48
1,497,674
h264
-7wwfGJXEZg_frame287_fps25.0
848
480
25
25/1
230
9.2
1,454,560
h264
-7wwfGJXEZg_frame300_fps25.0
848
480
25
25/1
300
12
3,428,785
h264
-Xg2-SUq6wo_frame339_fps25.0
848
480
25
25/1
331
13.24
2,443,698
h264
-dfvdKf-KR0__000001634800_000001643934
848
480
23.976024
24000/1001
220
9.175833
1,810,959
h264
-dfvdKf-KR0__000002977766_000002984273
848
480
23.976024
24000/1001
157
6.548208
744,208
h264
-eaFD5Nr2nw_frame240_fps25.0
848
480
25
25/1
240
9.6
1,903,735
h264
-mt8aNy1M00_frame280_fps24.0
848
480
23.976024
24000/1001
280
11.678333
1,866,090
h264
02jQiIkEGh8_frame148_fps24.0
848
480
23.976024
24000/1001
148
6.172833
1,838,328
h264
0BwnYerHb6Y_frame846_fps30.0
848
480
29.97003
30000/1001
846
28.2282
9,129,341
h264
0kRsiSdDFYg__000000116867_000000136928
848
480
23.976024
24000/1001
482
20.103417
7,677,008
h264
1R_kGaSJZ4o_frame207_fps30.0
848
480
30
30/1
207
6.9
3,961,402
h264
1ZGfUJML4Os_frame183_fps24.0
848
480
24
24/1
183
7.625
3,276,610
h264
1bk0rYvZXaI_frame639_fps30.0
848
480
29.97
2997/100
627
20.920921
3,639,249
h264
2W2ZkYARds4_frame890_fps30.0
848
480
29.97003
30000/1001
890
29.696333
4,697,067
h264
323v_FtWqvo__000000211344_000000218218
848
480
29.97003
30000/1001
207
6.9069
1,337,160
h264
3M4Saf-yiKY_frame147_fps24.0
848
480
23.976024
24000/1001
147
6.131125
2,755,716
h264
44G3dPME330__000000326451_000000333458
848
480
23.976024
24000/1001
164
6.840167
974,921
h264
5KLSf7kwr7s_frame182_fps30.0
848
480
29.97
2997/100
181
6.039373
1,519,940
h264
5XHF-kZftDA_frame187_fps24.0
848
480
23.976024
24000/1001
187
7.799458
2,179,690
h264
5Xm9P7bigo8_frame166_fps25.0
848
480
25
25/1
166
6.64
2,239,350
h264
62FAnfkm6vU_frame194_fps25.0
848
480
25
25/1
194
7.76
1,974,698
h264
6dmTmP4Dsw8__000000982315_000001005872
848
480
29.97003
30000/1001
708
23.6236
6,520,307
h264
6dmTmP4Dsw8__000001446779_000001473472
848
480
29.97003
30000/1001
801
26.7267
7,311,318
h264
6dmTmP4Dsw8__000002194325_000002207839
848
480
29.97003
30000/1001
406
13.546867
5,491,489
h264
74TEQfw6L60__000001066065_000001084717
848
480
29.97003
30000/1001
470
15.682333
2,059,143
h264
74TEQfw6L60__000001244710_000001253719
848
480
29.97003
30000/1001
270
9.009
1,533,295
h264
7E6i3E-fsj4__000000452367_000000461900
848
480
30
30/1
287
9.566667
2,915,819
h264
7MFKy7DJsCY__000000259559_000000270637
848
480
29.97003
30000/1001
333
11.1111
3,187,774
h264
8MkL3W6wU3g_frame666_fps30.0
848
480
29.97003
30000/1001
663
22.1221
2,918,272
h264
8UQFaVCV8S8__000000261261_000000275175
848
480
29.97
2997/100
418
13.947281
3,951,482
h264
8UxGzDeRIJk__000001448040_000001465280
848
480
25
25/1
431
17.24
5,689,390
h264
8UxGzDeRIJk__000001533160_000001550160
848
480
25
25/1
425
17
6,432,974
h264
8UxGzDeRIJk__000001876520_000001883760
848
480
25
25/1
181
7.24
2,262,505
h264
8hqkwCuJ0GU_frame265_fps30.0
848
480
29.97003
30000/1001
264
8.8088
1,798,762
h264
8znHT-ESDSA_frame200_fps24.0
848
480
23.976024
24000/1001
200
8.341667
1,327,077
h264
8znHT-ESDSA_frame303_fps24.0
848
480
23.976024
24000/1001
303
12.637625
2,050,016
h264
B7Hh0PY1kks__000002417280_000002426600
848
480
25
25/1
233
9.32
1,862,791
h264
BrjwEoPXyxA__000000724600_000000731800
848
480
30
30/1
216
7.2
1,680,004
h264
EP2rD6lr2Xk__000000929400_000000936840
848
480
25
25/1
186
7.44
2,229,765
h264
GDzuMTvHnJE__000000986120_000000995960
848
480
25
25/1
246
9.84
1,488,298
h264
HvKpnaXYUPU__000002823421_000002829627
848
480
29.97003
30000/1001
187
6.239567
1,470,059
h264
J54lSwAmxzQ_frame187_fps30.0
848
480
30
30/1
187
6.233333
3,877,012
h264
KWVSm6XHm7Q_frame195_fps30.0
848
480
29.97
2997/100
195
6.506507
1,269,168
h264
KWVSm6XHm7Q_frame231_fps30.0
848
480
29.97
2997/100
231
7.707708
1,846,164
h264
MQEcUPeAdFs__000000182482_000000191892
848
480
29.97003
30000/1001
283
9.442767
3,085,125
h264
aFfMGy94sjE__000000076376_000000083984
848
480
29.97003
30000/1001
164
5.472133
993,454
h264
aansXcMqnNk__000001956233_000001964133
848
480
30
30/1
217
7.233333
1,017,335
h264
aqFfjJrLkBA__000001289788_000001298063
848
480
29.97003
30000/1001
249
8.3083
2,266,108
h264
g8UiXbc19og__000000167120_000000173640
848
480
25
25/1
163
6.52
2,749,686
h264
gTPHH880Cyc__000000325640_000000340400
848
480
25
25/1
369
14.76
4,945,799
h264
mOiEOs3ZlT8__000000320654_000000328954
848
480
23.976024
24000/1001
201
8.383375
1,295,314
h264

LVO-Bench: Long Video Outpainting Benchmark

LongSCOP has been accepted at NeurIPS 2026.

LVO-Bench is the evaluation dataset introduced in LongSCOP: Semantically Consistent Long Video Outpainting by Yu Tang, Zhiyuan You, Zhixiao Wang, Fanghua Yu, Qingyu Zhang, Xiang Yin, Chao Dong, and Jinjin Gu (2026). It is designed to assess long-horizon generation stability, temporal coherence, and semantic consistency in video outpainting, with particular attention to methods that retrieve reference frames from across a video's timeline.

Source paper: arXiv:2611.12445 · Venue: NeurIPS 2026 (accepted) · Dataset Viewer

Video outpainting expands the visible spatial extent of a video while preserving its observed content. Over successive temporal chunks, a model must maintain consistent scene layouts, object identities, and appearances, while avoiding flicker, accumulated errors, and generation collapse.

This repository provides 52 full-frame evaluation videos and their technical metadata in a single test split. The 62K-clip training collection described in the LongSCOP paper is a separate dataset.

Benchmark construction

Section 4.1 of the LongSCOP paper describes a benchmark curated from LongVideoBench and Video-MME, with human expert curation and a multi-stage filtering pipeline:

  1. Shot segmentation: split source videos into temporally coherent shots and retain segments long enough to support temporally distant reference retrieval.
  2. Temporal variation filtering: remove near-static footage with little useful visual change, as well as clips with excessively violent motion.
  3. Geometric validation: use optical flow and robust point tracking to select clips with stable horizontal camera motion, limited rotation, and small scale drift.
  4. Reference recoverability: check that distant frames can align with target frames and reveal useful content near the left or right borders.

The resulting benchmark emphasizes videos in which reference retrieval can supply meaningful information for horizontal outpainting. Its selection favors recoverable geometry and should be considered when interpreting results on other motion patterns or unseen regions.

Release statistics

The following statistics are measured from the MP4 files in this release:

Property Value
Evaluation videos 52
Split test
Resolution 848 × 480 pixels, all videos
Container / video codec MP4 / H.264
Total video size 149,875,894 bytes (149.88 MB)
Total duration 581.90 seconds (9.70 minutes)
Clip duration 5.47–29.70 seconds; mean 11.19 seconds
Total decoded frames 16,011
Frames per clip 137–890
Frame rates Approximately 23.976–30 FPS; retained per video

num_frames is counted from decoded frames, and durations and frame rates are read from the encoded video streams. Filenames retain their original identifiers; frame-count or FPS strings embedded in a filename may describe an earlier processing stage. Use the metadata fields for the released files.

Dataset Viewer and fields

Open the Dataset Viewer to play videos and inspect their metadata. The dataset uses Hugging Face's native VideoFolder format.

Field Description
video Full-frame evaluation video, exposed as a Hugging Face Video feature.
video_id Unique clip identifier, matching the MP4 filename without its extension.
width, height Frame dimensions in pixels.
fps Numeric frame rate.
fps_rational Frame rate as an exact rational string, such as 30000/1001.
num_frames Number of decoded video frames.
duration_seconds Encoded video-stream duration in seconds.
size_bytes MP4 file size in bytes.
codec Video codec name.

The root-level metadata.jsonl contains one record per video. Its file_name field links each record to its MP4 and is converted into the video column by VideoFolder. This release contains full-frame videos and technical metadata; cropped inputs, masks, captions, and reference-frame annotations are not included.

Load with Hugging Face Datasets

Install a recent version of datasets with video support. Decoding frames also requires a compatible TorchCodec/PyTorch installation and FFmpeg; see the video loading guide.

from datasets import Video, load_dataset

dataset = load_dataset("tyxy/LVO-Bench", split="test")

# Inspect metadata and file paths without decoding video frames.
dataset = dataset.cast_column("video", Video(decode=False))
example = dataset[0]
print(example["video_id"], example["num_frames"], example["duration_seconds"])
print(example["video"]["path"])

To decode and sample a frame with a compatible video backend:

decoded = dataset.cast_column("video", Video())
frame = decoded[0]["video"].get_frame_at(0)
print(frame.data.shape)  # (channels, height, width)

Download the original files

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="tyxy/LVO-Bench",
    repo_type="dataset",
    local_dir="LVO-Bench",
    allow_patterns=["*.mp4", "metadata.jsonl", "README.md"],
)

Evaluation

Use these full-frame videos as targets and derive observed inputs and exterior masks under an explicitly documented outpainting protocol. Preserve temporal alignment between predictions and targets, and report the crop geometry, mask ratio, reference-selection policy, and frame sampling alongside results.

The LongSCOP paper reports PSNR and SSIM (higher is better), LPIPS and FVD (lower is better). Section 4.2 specifies frame values in [0, 1] for PSNR, SSIM, and FVD, and [-1, 1] for LPIPS. FVD uses 16 uniformly sampled frames per video. For reproducibility, also specify whether frame-level metrics cover the complete frame or only the generated region.

Reference and source attribution

The benchmark motivation and construction description above follow LongSCOP: Semantically Consistent Long Video Outpainting (arXiv:2611.12445), especially Sections 4.1–4.2. The numerical release statistics are measured from the files hosted here.

If you use LVO-Bench, please cite the LongSCOP paper:

@misc{tang2026longscop,
  title         = {LongSCOP: Semantically Consistent Long Video Outpainting},
  author        = {Yu Tang and Zhiyuan You and Zhixiao Wang and Fanghua Yu and Qingyu Zhang and Xiang Yin and Chao Dong and Jinjin Gu},
  year          = {2026},
  eprint        = {2611.12445},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  note          = {Accepted at NeurIPS 2026},
  url           = {https://arxiv.org/abs/2611.12445}
}

Please also acknowledge the source benchmarks, LongVideoBench and Video-MME.

License information

A standalone license for this release has not been specified. Source videos remain subject to their original rights and applicable source-dataset terms; this dataset card does not grant additional rights to the underlying footage.

Downloads last month
39

Papers for tyxy/LVO-Bench