LlamaSeg: Image Segmentation via Autoregressive Mask Generation

Published in Submitted to ICLR 2026, 2025

LlamaSeg reformulates segmentation as autoregressive visual generation, encoding masks as visual tokens for instruction-guided segmentation.

Recommended citation: Deng, Jiru, Tengjin Weng, Tianyu Yang, Wenhan Luo, Zhiheng Li, and Wenhao Jiang. (2025). "LlamaSeg: Image Segmentation via Autoregressive Mask Generation." Submitted to ICLR 2026.
Download Paper