Guardado en:
Detalles Bibliográficos
Autores principales: Wyrwiński, Piotr, Dobek, Kacper, Krawiec, Krzysztof
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2605.27696
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913170473353216
author Wyrwiński, Piotr
Dobek, Kacper
Krawiec, Krzysztof
author_facet Wyrwiński, Piotr
Dobek, Kacper
Krawiec, Krzysztof
contents Discrete visual tokenizers translate images into ordered sequences of codes, providing a natural representation for structural description of scenes. Yet existing adaptive tokenizers either require post-hoc search or select among a discrete set of pre-trained rates, rather than learning a continuous per-image sequence length coupled to the model and scene, and they typically train against pixel reconstruction, emphasizing texture rather than structure. We propose STROP, a discrete visual tokenizer architecture that forms structural scene representations and simultaneously learns how long an image's visual program should be. Using a four-phase curriculum supervised by local rate--distortion probes against frozen DINOv3 features, STROP optimizes a dedicated length head that estimates the active prefix length in a single forward pass. By bypassing pixel-level reconstruction gradients, the codebook is shaped entirely by the quality of higher-level latent representations. Program length grows with scene complexity, and signs of compositional structure emerge both in downstream dense-prediction transfer and in direct inspection of the learned code vocabulary.
format Preprint
id arxiv_https___arxiv_org_abs_2605_27696
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Structure over Pixels: Learning Variable-Length Visual Programs
Wyrwiński, Piotr
Dobek, Kacper
Krawiec, Krzysztof
Computer Vision and Pattern Recognition
Machine Learning
Discrete visual tokenizers translate images into ordered sequences of codes, providing a natural representation for structural description of scenes. Yet existing adaptive tokenizers either require post-hoc search or select among a discrete set of pre-trained rates, rather than learning a continuous per-image sequence length coupled to the model and scene, and they typically train against pixel reconstruction, emphasizing texture rather than structure. We propose STROP, a discrete visual tokenizer architecture that forms structural scene representations and simultaneously learns how long an image's visual program should be. Using a four-phase curriculum supervised by local rate--distortion probes against frozen DINOv3 features, STROP optimizes a dedicated length head that estimates the active prefix length in a single forward pass. By bypassing pixel-level reconstruction gradients, the codebook is shaped entirely by the quality of higher-level latent representations. Program length grows with scene complexity, and signs of compositional structure emerge both in downstream dense-prediction transfer and in direct inspection of the learned code vocabulary.
title Structure over Pixels: Learning Variable-Length Visual Programs
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2605.27696