Teaching an Agent to Sketch One Part at a Time

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Du, Xiaodan, Xu, Ruize, Yunis, David, Vinker, Yael, Shakhnarovich, Greg
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910160816963584
author Du, Xiaodan
Xu, Ruize
Yunis, David
Vinker, Yael
Shakhnarovich, Greg
author_facet Du, Xiaodan
Xu, Ruize
Yunis, David
Vinker, Yael
Shakhnarovich, Greg
contents We develop a method for producing vector sketches one part at a time. To do this, we train a multi-modal language model-based agent using a novel multi-turn process-reward reinforcement learning following supervised fine-tuning. Our approach is enabled by a new dataset we call ControlSketch-Part, containing rich part-level annotations for sketches, obtained using a novel, generic automatic annotation pipeline that segments vector sketches into semantic parts and assigns paths to parts with a structured multi-stage labeling process. Our results indicate that incorporating structured part-level data and providing agent with the visual feedback through the process enables interpretable, controllable, and locally editable text-to-vector sketch generation.
format Preprint
id arxiv_https___arxiv_org_abs_2603_19500
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Teaching an Agent to Sketch One Part at a Time
Du, Xiaodan
Xu, Ruize
Yunis, David
Vinker, Yael
Shakhnarovich, Greg
Artificial Intelligence
Computer Vision and Pattern Recognition
Graphics
Machine Learning
We develop a method for producing vector sketches one part at a time. To do this, we train a multi-modal language model-based agent using a novel multi-turn process-reward reinforcement learning following supervised fine-tuning. Our approach is enabled by a new dataset we call ControlSketch-Part, containing rich part-level annotations for sketches, obtained using a novel, generic automatic annotation pipeline that segments vector sketches into semantic parts and assigns paths to parts with a structured multi-stage labeling process. Our results indicate that incorporating structured part-level data and providing agent with the visual feedback through the process enables interpretable, controllable, and locally editable text-to-vector sketch generation.
title Teaching an Agent to Sketch One Part at a Time
topic Artificial Intelligence
Computer Vision and Pattern Recognition
Graphics
Machine Learning
url https://arxiv.org/abs/2603.19500