WeVisOne-VL paper: unified object-to-pixel visual perception on arXiv
lmoroney · x · 2026-10-26
The WeVisOne-VL paper (arXiv:2610.31329) details a VLM that unifies object-level and pixel-level perception via structured next-token prediction, supporting detection, referring localization, counting, OCR, segmentation, and matting at any granularity from points/boxes to masks and alpha mattes.
- Architecture: a lightweight pixel decoder is integrated into the VLM via a special <OBJ> token for instance-level pixel outputs
- Training: three-stage curriculum SFT followed by mixed-task RL
- Results: leading performance across 30+ benchmarks on both object- and pixel-level tasks, with robust rejection of object-absent queries
- Open-sourced; the goal is moving VLMs from broad understanding toward reliable perception for real-world deployment
More from Models
- I Asked 5 Text-Only LLMs to Draw Objects They've Never Seen — gabriberton · 2026-10-26
- Japan to build AI cyberattack simulation team as AI defense goes mainstream — evijit · 2026-10-26
- Anthropic engineer: Claude Code cache hit rates are now high, savings live — teortaxesTex · 2026-10-26
- Fine-tuned 2015 CNN beats GPT-6 and Claude on skin lesion diagnosis, 62% vs 42% — lmoroney · 2026-10-26
- "Never seen Claude this happy": community speculates Opus 4.5 is imminent — adonis_singh · 2026-10-26
- 'Absolute humiliation': China open-weights model outscoring OpenAI flagship draws mockery — alejandroll10 · 2026-10-26