LlamaSeg: Image Segmentation via Autoregressive Mask Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Deng, Jiru, Weng, Tengjin, Yang, Tianyu, Luo, Wenhan, Li, Zhiheng, Jiang, Wenhao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913859342696448
author Deng, Jiru
Weng, Tengjin
Yang, Tianyu
Luo, Wenhan
Li, Zhiheng
Jiang, Wenhao
author_facet Deng, Jiru
Weng, Tengjin
Yang, Tianyu
Luo, Wenhan
Li, Zhiheng
Jiang, Wenhao
contents We present LlamaSeg, a visual autoregressive framework that unifies multiple image segmentation tasks via natural language instructions. We reformulate image segmentation as a visual generation problem, representing masks as "visual" tokens and employing a LLaMA-style Transformer to predict them directly from image inputs. By adhering to the next-token prediction paradigm, our approach naturally integrates segmentation tasks into autoregressive architectures. To support large-scale training, we introduce a data annotation pipeline and construct the SA-OVRS dataset, which contains 2M segmentation masks annotated with over 5,800 open-vocabulary labels or diverse textual descriptions, covering a wide spectrum of real-world scenarios. This enables our model to localize objects in images based on text prompts and to generate fine-grained masks. To more accurately evaluate the quality of masks produced by visual generative models, we further propose a composite metric that combines Intersection over Union (IoU) with Average Hausdorff Distance (AHD), offering a more precise assessment of contour fidelity. Experimental results demonstrate that our method surpasses existing generative models across multiple datasets and yields more detailed segmentation masks.
format Preprint
id arxiv_https___arxiv_org_abs_2505_19422
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LlamaSeg: Image Segmentation via Autoregressive Mask Generation
Deng, Jiru
Weng, Tengjin
Yang, Tianyu
Luo, Wenhan
Li, Zhiheng
Jiang, Wenhao
Computer Vision and Pattern Recognition
We present LlamaSeg, a visual autoregressive framework that unifies multiple image segmentation tasks via natural language instructions. We reformulate image segmentation as a visual generation problem, representing masks as "visual" tokens and employing a LLaMA-style Transformer to predict them directly from image inputs. By adhering to the next-token prediction paradigm, our approach naturally integrates segmentation tasks into autoregressive architectures. To support large-scale training, we introduce a data annotation pipeline and construct the SA-OVRS dataset, which contains 2M segmentation masks annotated with over 5,800 open-vocabulary labels or diverse textual descriptions, covering a wide spectrum of real-world scenarios. This enables our model to localize objects in images based on text prompts and to generate fine-grained masks. To more accurately evaluate the quality of masks produced by visual generative models, we further propose a composite metric that combines Intersection over Union (IoU) with Average Hausdorff Distance (AHD), offering a more precise assessment of contour fidelity. Experimental results demonstrate that our method surpasses existing generative models across multiple datasets and yields more detailed segmentation masks.
title LlamaSeg: Image Segmentation via Autoregressive Mask Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2505.19422