README.md

metadata

configs:
  - config_name: default
    data_files:
      - path: data/**/*.parquet
        split: train
license: odc-by

Molmo2-VideoTrack

Molmo2-VideoTrack is a dataset of video point tracking annotations collected from human annotators across 16 video datasets. It can be used to fine-tune vision-language models for video object tracking via point trajectories.

Molmo2-VideoTrack is part of the Molmo2 dataset collection and was used to train the Molmo2 family of models.

Quick links:

• 📃 Paper
• 🎥 Blog with Videos

Usage

from datasets import load_dataset

# Load entire dataset
ds = load_dataset("allenai/Molmo2-VideoTrack", split="train")

# Filter by video dataset
dancetrack = ds.filter(lambda x: x == 'dancetrack', input_columns='video_dataset')

Data Format

Each row contains tracking annotations for one or more objects in a video clip:

Field	Description
id	Unique identifier for this annotation
video	Video filename
clip	trimmed clip id
video_dataset	Source dataset name (e.g., 'dancetrack', 'mose')
video_source	Video directory used in training (can be ignored)
exp	Text expression describing the tracked object(s)
obj_id	List of object IDs per video
mask_id	List of mask IDs corresponding to tracked objects starting from '0'
points	List of point trajectories per object. Each entry contains object_id (corresponding to an ID in mask_id) and points (list of [x, y] coordinates per frame). Example: [{'object_id': '0', 'points': [[x1, y1], [x2, y2], ...]}, ...]
segments	List of segment annotations per object. Each entry contains object_id (corresponding to an ID in mask_id) and segments. Example: [{'object_id': '0', 'segments': [...]}, ...]
start_frame	Starting frame index for this clip (use to trim the source video)
end_frame	Ending frame index for this clip (use to trim the source video)
w	Video width
h	Video height
n_frames	Number of frames in the clip
fps	Used in training

Important: start_frame and end_frame indicate which portion of the source video to use. You need to trim the video to this range — the annotations correspond to frames within [start_frame, end_frame], not the entire video.

Folder Structure

Molmo2-VideoTrack/
├── README.md
└── data/
    ├── animaltrack/
    │   └── point_tracks.parquet
    ├── APTv2/
    │   └── point_tracks.parquet
    ├── ...
    └── {video_dataset}/
        └── point_tracks.parquet

Video Sources

The table below contains information on the sources of the third party datasets used or referenced in curating the data for Molmo2-VideoTrack. We do not provide video files or share the original raw data from datasets with restrictions on use and distribution according to the source license. We instead provide the links, license information, and notes for downloading videos from the original datasets for transparency and reproducibility. Please verify the licenses and use requirements that apply to each dataset before downloading as they may change or be updated by the dataset providers.

Dataset	Category	Annotation Source	Download	Dataset License	Note
mose	General	Segmentation	MOSE	CC BY-NC-SA 4.0
mosev2	General	Segmentation	MOSEv2	CC BY-NC-SA 4.0
sav	General	Segmentation	SA-V	CC BY 4.0	Sampled at 6 fps from the original 24 fps video to match the segmentation annotation
vipseg	General	Segmentation	VIPSeg	Non-commercial research use only	Change to 720p format
animaltrack	Animals	Bounding Box	AnimalTrack	Non-commercial research use only	Train and val videos are used due to data scarcity
APTv2	Animals	Bounding Box	APTv2	Apache 2.0
bft	Bird Flocks	Bounding Box	BFT	Apache 2.0
soccernet	Sports	Bounding Box	SoccerNet	Non-commercial research use only	Fill in the NDA form to access the videos
sportsmot	Sports	Bounding Box	SportsMOT	CC BY-NC 4.0
teamtrack	Sports	Bounding Box	TeamTrack	MIT
mot2020	Pedestrians	Bounding Box	MOT20	CC BY-NC-SA 3.0
personpath22	Pedestrians	Bounding Box	PersonPath22	CC BY-NC 4.0
dancetrack	Dancers	Bounding Box	DanceTrack	Non-commercial research use only
bdd100k	Autonomous Driving	Bounding Box	BDD100K	BSD-3	Download only bdd100k_videos_train_00.zip
uavdt	UAV	Bounding Box	UAVDT	Research use only
seadrones	UAV	Bounding Box	SeaDronesSee	CC0 / Unknown	Use 'Multi-Object Tracking'

License

This dataset is licensed under ODC-BY-1.0. It is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines. Please refer to the Video Sources section for the original datasets that provide the videos used to generate the segmentations and point tracks for this dataset. All use of the videos and original data from these datasets are subject to the licenses and terms of use provided by the sources. Please check the sources to determine if they are appropriate for your use case.