AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

LlamaSeg: Image Segmentation via Autoregressive Mask Generation

Jiru Deng, Tengjin Weng, Tianyu Yang, Wenhan Luo, Zhiheng Li, Wenhao Jiang

ECCV 2026 · Detection and segmentation › Open-set and few-shot detection · approach: Open-vocabulary semantic segmentation

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
increase mask token resolution to 384 pix.LlamaSeg-L · ADE20KmIoU: 52.0 -> 55.9Improved
increase mask token resolution to 384 pix.LlamaSeg-L · COCO-StuffmIoU: 55.7 -> 58.1Improved
increase mask token resolution to 384 pix.LlamaSeg-L · PAS-20mIoU: 75.1 -> 77.1Improved
add SA-OVRS plus referring-data pre-training (ablation)LlamaSeg-B · ADE20KmIoU: 48.9 -> 50.5Improved
add SA-OVRS plus referring-data pre-training (ablation)LlamaSeg-B · refCOCOcIoU: 46.0 -> 50.9Improved
fine-tune VQGAN tokenizer on masks for 5000 steps (ablation)VQGAN mask tokenizer · mask reconstructiontotal IoU: 96.8 -> 93.0Hurt
fine-tune VQGAN tokenizer on masks for 10000 steps (ablation)VQGAN mask tokenizer · mask reconstructionmAHD: 6.1 -> 10.6Hurt
use greedy search instead of beam search (B=3) (ablation)LlamaSeg-B · RefCOCO valcIoU: 47.3 -> 50.9Improved
use greedy search instead of random sampling (ablation)LlamaSeg-B · RefCOCO valcIoU: 49.2 -> 50.9Improved
use greedy search instead of Top-P (P=0.9) (ablation)LlamaSeg-B · RefCOCO valcIoU: 50.2 -> 50.9Improved
scale LlamaSeg from B to LLlamaSeg · refCOCO valcIoU: 50.9 -> 56.5Improved
scale LlamaSeg from B to LLlamaSeg · refCOCO+ valcIoU: 38.9 -> 41.6Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy