FlexTok: Resampling Images into 1D Token Sequences of Flexible Length
Roman Bachmann, Jesse Allardice, David Mizrahi, Enrico Fini, Oğuzhan Fatih Kar, Elmira Amirloo, Alaaeldin El-Nouby, Amir Zamir, Afshin Dehghan · ICML 2025 · 2025
- Status
- To read
- Topics
- budgeted-inference, other
- Feeds
- P2 · NeurIPS 2027
What they did
Resamples 2D images into ordered 1D token sequences of flexible length, so the number of tokens can follow the complexity of the image.
My note
Uses ordered, variable-length representations and nested dropout. Useful inspiration for nested sampling budgets.
Code
- Repository
- apple-aiml-research/ml-flextok
- Framework
- PyTorch
- Licence
- See the repository
- My status
- Not tried
BibTeX
@inproceedings{bachmann2025-flextok,
title = {{FlexTok: Resampling Images into 1D Token Sequences of Flexible Length}},
author = {Roman Bachmann and Jesse Allardice and David Mizrahi and Enrico Fini and Oğuzhan Fatih Kar and Elmira Amirloo and Alaaeldin El-Nouby and Amir Zamir and Afshin Dehghan},
booktitle = {ICML 2025},
year = {2025},
eprint = {2502.13967},
archivePrefix = {arXiv}
}