google-deepmind / google-deepmind/tapnet

License of the TAPNext++ checkpoint files appears unstated

Open
#164 3 comments 0 reactions 0 assignees View on GitHub
Dominant language
Jupyter Notebook
Stars
2k
Forks
192
PR merge metrics
No merged PRs in 30d

Description

Thank you for releasing TAPNext++ — the online tracking and re-detection
behavior is very impressive.

We are evaluating the released checkpoints for use in a downstream project
and could not find a license statement that covers the checkpoint files
themselves:

- `https://storage.googleapis.com/dm-tapnet/tapnextpp/tapnextpp_ckpt.pt`
(linked from `tapnet/tapnextpp/README.md`)
- `https://storage.googleapis.com/gresearch/tapnextpp/tapnextpp_512.ckpt`
(linked from `tapnet/tapnextpp/votsp2026/README.md`)

Places we checked:

- The repository `LICENSE` and the README's "License and Disclaimer"
section: they cover the software (Apache 2.0) and the TAP-Vid
non-software materials (TAP-Vid annotations, RGB-Stacking and RoboTAP
videos — CC-BY 4.0), but we observed no statement covering released
model checkpoints.
- `tapnet/tapnextpp/README.md` and
`tapnet/tapnextpp/votsp2026/README.md`: download instructions only, no
license statement for the weights.
- The TAPNext++ project page: no license statement for the weights.
- The GCS object metadata for both files: no license metadata.
- The HuggingFace mirror https://huggingface.co/google/tapnet: tagged
`apache-2.0` and contains TAPIR / BootsTAPIR / BootsTAPNext
checkpoints, but the TAPNext++ files are not included there.

Could you clarify whether the TAPNext++ checkpoints are released under
Apache 2.0 like the code (and like the checkpoints mirrored on the
HuggingFace repo), or under different terms? A one-line statement in
`tapnet/tapnextpp/README.md` — or adding the files to the HuggingFace
repo where the license tag applies — would resolve the ambiguity for
downstream users.

Thanks!

Contributor guide

Open the contributing guide

Research direction

Read tapnet/tapnextpp/README.md, tapnet/tapnextpp/votsp2026/README.md, and the repository LICENSE, then compare the stated terms with the two checkpoint download links. Confirm the applicable checkpoint license with the maintainers; done means both README guidance and the downstream usage terms are unambiguous.

Written by the indexing model from the issue text.

Assessment

Tech stack
huggingface, machine-learning
Domain
documentation, machine-learning
Issue type
Documentation
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.