ECCV 2026
UniReflect: Self-Reflection Tuning for Unified Multimodal Understanding and Generation
First unified visual understanding-and-generation framework with in-model self-reflection verification.
I am a researcher at Meituan Inc., where I work on large multimodal models, diffusion models, and AI design agents.
My research is particularly focused on building commercial image generation systems — spanning the full stack from data curation, training of large-scale generative models, to rigorous evaluation of generation quality.
Previously, my work mainly explored the application of multimodal large language models to fine-grained visual perception — including language-guided segmentation, multi-granular visual understanding, and self-reflection frameworks for multimodal understanding and generation.
First unified visual understanding-and-generation framework with in-model self-reflection verification.
Plug-and-play video tokenizer preserving image–language alignment for image-to-video adaptation.
Open-source bilingual image-generation foundation model with state-of-the-art multilingual text rendering.
First VLLM-based universal segmentation model for pixel-level image and video perception.
End-to-end MLLM unifying referring and reasoning segmentation across image and video.
Multi-granular perception framework unifying word- and sentence-level tasks with box and mask predictions.