Dataset Viewer
Auto-converted to Parquet Duplicate
sequence_id
string
category
string
bboxes
list
full_occlusion
list
out_of_view
list
visible
list
description
string
frame_names
list
num_frames
int32
width
int32
height
int32
video
video
video_width
int32
video_height
int32
fps
float64
airplane-2
airplane
[[267.0,261.0,816.0,228.0],[262.0,260.0,821.0,229.0],[263.0,258.0,820.0,230.0],[263.0,257.0,821.0,23(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
brown airplane flying in the air
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
8,301
1,280
720
1,280
720
30
airplane-3
airplane
[[477.0,180.0,581.0,155.0],[477.0,178.0,580.0,154.0],[477.0,173.0,580.0,163.0],[477.0,173.0,581.0,16(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
airplane colored in white and red flying above the grass
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
2,040
1,280
720
1,280
720
30
airplane-4
airplane
[[615.0,367.0,244.0,68.0],[608.0,361.0,250.0,72.0],[608.0,359.0,246.0,75.0],[607.0,358.0,252.0,74.0](...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
airplane colored in white and green flying in the air
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
5,141
1,280
720
1,280
720
30
airplane-5
airplane
[[496.0,368.0,221.0,31.0],[496.0,367.0,228.0,36.0],[498.0,367.0,232.0,37.0],[503.0,367.0,232.0,37.0](...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
airplane flying among other airplanes
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
5,923
1,280
720
1,280
720
30
airplane-6
airplane
[[165.0,295.0,30.0,14.0],[164.0,292.0,30.0,14.0],[164.0,291.0,32.0,14.0],[163.0,289.0,33.0,16.0],[16(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
airplane flying in the air
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
1,101
406
720
406
720
30
airplane-7
airplane
[[182.0,6.0,124.0,153.0],[181.0,6.0,124.0,153.0],[182.0,6.0,123.0,152.0],[186.0,9.0,123.0,152.0],[18(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
airplane flying above the road
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
2,888
640
360
640
360
30
airplane-8
airplane
[[629.0,349.0,33.0,22.0],[627.0,355.0,34.0,15.0],[627.0,355.0,33.0,15.0],[625.0,355.0,34.0,16.0],[62(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
red airplane flying in the air
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
7,201
1,280
720
1,280
720
30
airplane-10
airplane
[[610.0,341.0,136.0,28.0],[611.0,342.0,136.0,28.0],[606.0,344.0,144.0,27.0],[604.0,345.0,146.0,26.0](...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
airplane flying in the air
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
1,568
1,280
720
1,280
720
30
airplane-11
airplane
[[604.0,286.0,95.0,43.0],[603.0,287.0,97.0,39.0],[604.0,287.0,97.0,39.0],[604.0,288.0,98.0,39.0],[60(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
airplane colored in yellow and red flying in the air
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
5,016
1,280
720
1,280
720
30
airplane-12
airplane
[[632.0,258.0,133.0,119.0],[627.0,252.0,137.0,123.0],[627.0,250.0,137.0,122.0],[627.0,247.0,137.0,12(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED)
[true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED)
white airplane flying in the air above the forest
["00000001.jpg","00000002.jpg","00000003.jpg","00000004.jpg","00000005.jpg","00000006.jpg","00000007(...TRUNCATED)
2,660
1,280
720
1,280
720
30
End of preview. Expand in Data Studio

LaSOT — sequence-level MP4 videos and bounding boxes

Status: INCOMPLETE — conversion/upload in progress. Uploaded: 384 train and 96 test sequences across 24 categories. The full conference benchmark has 1,120 train / 280 test sequences, 70 categories, and more than 3.5 million frames. The separate 150-sequence journal extension is not included.

Source: LaSOT. Original archives were obtained from l-lt/LaSOT at revision a97464600e8c1ab91eb91a47332a20732830585e. Each downloaded ZIP is verified against the SHA-256 and byte size in that pinned mirror's Git LFS metadata. Sequence lengths and dimensions are checked against the original website metadata. The mirror ZIPs may differ from the authors' archived ZIPs; an original-author checksum match is not claimed. Actual SHA-1 hashes and official checksum match results are recorded per category in conversion_manifest.json; original category checksums are retained for comparison. For a local-source conversion, archive checksums are not applicable; see the manifest.

Load

from datasets import load_dataset, Video

ds = load_dataset("ghomasHudson/lasot", split="test", streaming=True)
# Metadata-only access does not require torchcodec or download the video files.
row = next(iter(ds.cast_column("video", Video(decode=False))))
print(row["video"]["path"], row["bboxes"][0])

# With compatible torch, torchcodec and FFmpeg installed:
row = next(iter(ds))
frame = row["video"].get_frame_at(0).data
bbox = row["bboxes"][0]  # [x, y, width, height] for that exact frame

Videos are separate files in this repository, referenced by the Video column with hf:// paths. Parquet shards contain metadata, not embedded MP4 bytes. Loading metadata does not download every video; decoding fetches the selected video on demand. For a completely offline copy, download the repository and rewrite video.path to the corresponding local MP4 before decoding. A Parquet-only download is not sufficient.

Columns and alignment

One row is one complete, untrimmed sequence. All JPEGs appear once in numeric order. No temporal subsampling, interpolation, frame duplication, or audio is added.

Column Meaning
sequence_id, category Original sequence ID and category
video Hugging Face Video feature pointing to the sequence's MP4
bboxes Per-frame [x, y, width, height], absolute pixels, original values/origin
full_occlusion, out_of_view Original per-frame flags; true means occluded/out of view
visible not (full_occlusion or out_of_view); not a partial-occlusion label
description Original nlp.txt language description
frame_names Original JPEG filenames, aligned with video and annotations
num_frames Number of video frames and entries in every per-frame list
width, height Original JPEG dimensions
video_width, video_height Encoded dimensions, including any right/bottom padding
fps Container playback rate: 30.0 FPS

bboxes[i], each flag at index i, and frame_names[i] correspond to decoded video frame i (zero-based list indexing). Bounding-box coordinates are not normalized, clipped, shifted to a new origin, or changed to xyxy. All-zero absent boxes are retained. Consumers must apply any coordinate-origin conversion explicitly. The playback rate is a conversion setting, not a claim about each source camera's FPS.

Encoding and validation

Lossy H.264 (libx264, CRF 18, yuv420p). This is a re-encoded representation, not a byte-exact copy of the original JPEG frames. Odd image dimensions are padded by one pixel on the right/bottom, never resized. Preset: veryfast. No spatial resizing. Encoding settings and completed categories are recorded in conversion_manifest.json. Every output is decoded by ffprobe to check its frame count; annotation lengths and image dimensions are validated first. The official Protocol II split is used; there is no invented validation split. Protocol I evaluates the union of train and test.

Terms and attribution

The original download page states: “LaSOT can be freely used for personal, academic, or educational purposes.” Copyright remains with the LaSOT authors and the respective source-video owners. This conversion does not grant additional rights or assert a permissive software license over the data. Check the original terms before redistribution or other uses.

@inproceedings{fan2019lasot,
  title={LaSOT: A High-quality Benchmark for Large-scale Single Object Tracking},
  author={Fan, Heng and Lin, Liting and Yang, Fan and Chu, Peng and Deng, Ge and
          Yu, Sijia and Bai, Hexin and Xu, Yong and Liao, Chunyuan and Ling, Haibin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2019}
}
Downloads last month
470