Yong Xien Chng

Research

Selected work is highlighted.

The same prompt rendered at loop depth 1 to 4; the text becomes correct by loop 4.

Looped Diffusion Transformer

Yong Xien Chng, et al.

arXiv preprint, 2026 (under review)

arXiv soonCode soon

Rerun the middle blocks of a diffusion transformer inside every denoising step. A 260M model beats models 6.5× larger at 1/5 the compute, and later loops correct mistakes made by earlier ones.

InterSketch benchmark results and examples of interleaved visual reasoning.

InterSketch: An Interleaved Reasoning Model with Self-correcting Visual Sketch and Stepwise Reward

Zhiwei Ning, Wenwen Tong, Xiangli Kong, Shengnan Ma, Yong Xien Chng, et al.

arXiv preprint, 2026

arXiv

A vision-language model that thinks by drawing. It sketches intermediate figures with tools, reflects on them, and is trained with stepwise rewards so RL can reach the end of long trajectories.

Semantic segmentation examples comparing predictions with ground truth.

Aligning Vision-Language Model with Fine-Grained Semantics for Open-Vocabulary Segmentation

Yong Xien Chng, Xuchong Qiu, Yizeng Han, Kai Ding, Wan Ding, Gao Huang

IEEE Transactions on Circuits and Systems for Video Technology, 2026

arXiv

The bottleneck in open-vocabulary segmentation is classifying masks, not finding them. Semantic-guided mask attention and tuning only the query projection fix it cheaply.

SenseNova-MARS: a policy VLM interleaving image search, text search and image crop tools with reasoning.

SenseNova-MARS: Empowering Multimodal Agentic Reasoning and Search via Reinforcement Learning

Yong Xien Chng*, Tao Hu*, Wenwen Tong*, Xueheng Li, Jiandong Chen, Haojia Yu, Jiefan Lu, Hewei Guo, Hanming Deng, Chengjun Xie, Gao Huang, Dahua Lin, Lewei Lu (* equal contribution)

CVPR, 2026 (Highlight)

CVPR paper arXiv Code Models

A VLM that searches the web, crops the image and reasons, all in one interleaved loop learned with RL. The 32B model beats Gemini-3-Pro and GPT-5.2 on multimodal search, and we release HR-MMSearch, a high-resolution search benchmark.

A 1000-page PDF exceeds a model's context; with multimodal RAG the model retrieves the right pages and answers.

Scaling Beyond Context: A Survey of Multimodal Retrieval-Augmented Generation for Document Understanding

Sensen Gao, Shanshan Zhao, Xu Jiang, Lunhao Duan, Yong Xien Chng, et al.

ACL, 2026

arXiv Project Paper list

A map of retrieval-augmented generation for documents that are more than text: tables, charts, layout, and the agentic pipelines that read them.

Diagram of a diffusion-based unified model: text tokenizer and pixel encoder feeding a diffusion backbone and pixel decoder.

Unified Multimodal Understanding and Generation Models: Advances, Challenges, and Opportunities

Shanshan Zhao*, Xinjie Zhang*, Jintao Guo*, Jiakui Hu, Lunhao Duan, Minghao Fu, Yong Xien Chng, Guo-Hua Wang, Qing-Guo Chen, Zhao Xu, Weihua Luo, Kaifu Zhang (* equal contribution)

arXiv preprint, 2025

arXiv Paper list

How understanding and generation are being folded into a single model, organised by architecture (diffusion, autoregressive, hybrid), and what still keeps them apart.

DenseGrounding: a 3D scan with a box grounded next to a printer, guided by an LLM-augmented description.

DenseGrounding: Improving Dense Language-Vision Semantics for Ego-Centric 3D Visual Grounding

Henry Zheng*, Hao Shi*, Qihang Peng, Yong Xien Chng, Rui Huang, Yepeng Weng, Zhongchao Shi, Gao Huang (* equal contribution)

ICLR, 2025

arXiv OpenReview

Grounding language in 3D from multi-view images needs dense scene semantics and richer language. Won 1st place and the Innovation Award at the CVPR 2024 Autonomous Grand Challenge.

Skywork R1V benchmark bars on MMMU and PhyX against GPT-4.5, GPT-4o and Claude 3.7 Sonnet.

Skywork R1V: Pioneering Multimodal Reasoning with Chain-of-Thought

Yi Peng, Peiyu Wang, Xiaokun Wang, Yichen Wei, Jiangbo Pei, Weijie Qiu, Ai Jian, Yunzhuo Hao, Jiachun Pan, Tianyidan Xie, Li Ge, Rongxian Zhuang, Xuchen Song, Yang Liu, Yahui Zhou

Technical report, 2025

arXiv Code Model

Bringing R1-style chain-of-thought to vision without retraining the LLM or the vision encoder: a light projector, iterative SFT with GRPO, and distilled reasoning chains that know when to stop.

A 3D indoor point cloud segmented into semantic classes by Meepo.

Exploring Contextual Modeling with Linear Complexity for Point Cloud Segmentation

Yong Xien Chng, Xuchong Qiu, Yizeng Han, Yifan Pu, Jiewei Cao, Gao Huang

arXiv preprint, 2024

arXiv

Mamba works for point clouds once you drop causality and scan in both directions. Beats PTv3 while being 42% faster and 5.5× more memory-efficient.

1^stChallenge winnerCVPR 2024 · 3D Grounding

DenSeg: Alleviating Vision-Language Feature Sparsity in Multi-View 3D Visual Grounding

Henry Zheng, Hao Shi, Yong Xien Chng, Rui Huang, Zanlin Ni, Tianyi Tan, Qihang Peng, Yepeng Weng, Zhongchao Shi, Gao Huang

CVPR 2024 Autonomous Grand Challenge Workshop

1st place, Multi-view 3D Visual Grounding track. The challenge entry that grew into DenseGrounding.

Mask Grounding: a referring expression with one word masked, matched to the object it names.

Mask Grounding for Referring Image Segmentation

Yong Xien Chng, Henry Zheng, Yizeng Han, Xuchong Qiu, Gao Huang

CVPR, 2024

arXiv Code

Teach the model which word refers to which object, and referring segmentation gets better. A simple auxiliary task that lifts prior methods and sets the state of the art on RefCOCO, RefCOCO+ and G-Ref.