Ant Group proposes ARGenSeg-8B: an image segmentation framework built on autoregressive image generation models
Original title:inclusionAI/ARGenSeg-8B
AI summary
Ant Group introduced ARGenSeg-8B, an autoregressive image generation segmentation framework that unifies multimodal understanding and pixel-level perception. It uses a multimodal large language model (MLLM) to output visual tokens and decodes them into segmentation masks via a general VQ-VAE, making segmentation rely entirely on the MLLM's pixel-level understanding. It adopts a next-scale-prediction strategy to generate visual tokens in parallel, reducing inference latency. It outperforms previous state-of-the-art methods on multiple segmentation datasets with significantly faster inference. The paper has been accepted to NeurIPS 2025, and the model has been released on HuggingFace.
Key Takeaways
- Ant Group introduced ARGenSeg-8B,an autoregressive image generation segmentation framework that…
- It uses a multimodal large language model (MLLM) to output visual tokens and decodes them into…
- It adopts a next-scale-prediction strategy to generate visual tokens in parallel,reducing…
- It outperforms previous state-of-the-art methods on multiple segmentation datasets with…
- The paper has been accepted to NeurIPS 2025,and the model has been released on HuggingFace.
Quick Facts
- Published
- Updated
- Category
- Model Releases
- AI score
- 56
蚂蚁提出用自回归生成做分割,把理解和像素级感知统一到一个框架里,多个数据集SOTA且速度更快,做CV的值得看看。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Ant Group proposes ARGenSeg-8B: an image segmentation framework built on autoregressive image generation models. May 15, 2026. https://new.nat6.net/en/news/cmp6c0i6i00nbslnzm1hefdiu/