UMM-Reflection open-sourced: interleaved RL teaches unified models to self-correct images, GenEval 0.71→0.84
ziqi_huang_ · x · 2026-09-29
Ziqi Huang and collaborators open-sourced UMM-Reflection, a method that trains unified multimodal models to natively generate, reflect, and redraw via interleaved reinforcement learning: the model diagnoses errors in its own image, revises it, and iterates, with reflection text and image generation learned jointly inside one model.
How it works
- RL over complete reflection trajectories: sibling trajectories share one initial image, and a group-relative advantage compares reflection strategies; a trajectory-level advantage updates both reflection tokens and flow-based revisions, avoiding combinatorial blow-up of per-round credit assignment.
- Credit flows across rounds and across both roles of the same model; no external verifier needed at inference.
- SFT on reflection trajectories gives only a cold start; RL is what finds high-success repair paths.
Results: on BAGEL, GenEval improves 12.05 points over SFT (0.71 → 0.84), with transfer gains on unseen benchmarks: WISE (+10.97), T2I-CompBench++ (+4.63), and OneIG-Bench (+3.48).
Paper, code, models, and demo are all released.
More from Research
- Open model Inkling 2 Small repairs broken networks from raw weights in ~40 tokens — SonglinYang4 · 2026-11-14
- Harvard-MIT UGG model unifies video generation, depth and camera pose in one diffusion model — burny_tech · 2026-11-14
- Hugging Face Introduces Decision Models That Output Probability Distributions Without Generating Text — ariG23498 · 2026-11-14
- New results in computable topology reduce computable structures to two theorems — jessi_cata · 2026-11-14
- BAAI's encoder-free VLM family EVE: NeurIPS spotlight, ICCV highlight, open-sourced — liuziwei7 · 2026-11-14
- Recursive Model Collapse: How It Learned to Stop Worrying and Become the Assistant — sebkrier · 2026-11-14