RAR: Rate-Adaptive Rollout for Time-Consistent Interactive World Models
Yaoyang Liu*, Dongyu Yan*, Yihua Du, Shengju Qian, Zeke Xie, Xin Wang, Ying-Cong Chen, Long Chen.
arXiv preprint (2026)
Arxiv / Paper / Github / Project Page

We propose Rate-Adaptive Rollout (RAR), which decouples the temporal rate of interactive world models from generation throughput. Stride-RoPE and synchronized multi-stride sampling enable time-consistent rollout at different rates, including unseen rates and online rate switching, while extending physical horizons.

UltraShape 1.0: High-Fidelity 3D Shape Generation via Scalable Geometric Refinement
Tanghui Jia*, Dongyu Yan*, Dehao Hao*, Yang Li, Kaiyi Zhang, Xianyi He, Lanjiong Li, Yuhan Wang, Jinnan Chen, Lutao Jiang, Qishen Yin, Long Quan, Ying-Cong Chen, Li Yuan.
arXiv preprint (2025)
Arxiv / Paper / Github / Project Page

We present a scalable two-stage 3D diffusion framework that first synthesizes coarse geometry and then refines it into detailed, high-fidelity shapes. A geometry-aware refinement approach and a watertight data-processing pipeline preserve fine geometric details and improve generation quality.

FlexPainter: Flexible and Multi-View Consistent Texture Generation
Dongyu Yan*, Leyi Wu*, Jiantao Lin, Luozhou Wang, Tianshuo Xu, Zhifei Chen, Zhen Yang, Lie Xu, Shunsi Zhang, Ying-Cong Chen.
arXiv preprint (2025)
Arxiv / Paper / Github / Project Page

We introduce FlexPainter, a flexible texture generation pipeline supporting text, image, and combined multimodal guidance. Joint multi-view generation, view synchronization, and 3D-aware texture completion produce consistent, high-resolution textures for 3D assets.

MSI-NeRF: Linking Omni-Depth with View Synthesis through Multi-Sphere Image aided Generalizable Neural Radiance Field
Dongyu Yan, Guanyu Huang, Fengyu Quan, Haoyao Chen.
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2025)
Arxiv / Paper / Video / Github

We create a generalizable NeRF that takes four fisheye images as input and outputs a 3D MSI representation for novel view synthesis and depth estimation. It can be trained with synthetic depth data only and can generalize to a wide range of scenarios. We also released a fisheye multi-view dataset for training and evaluation.

Active Implicit Object Reconstruction using Uncertainty-guided Next-Best-View Optimization
Dongyu Yan*, Jianheng Liu*, Fengyu Quan, Haoyao Chen.
IEEE Robotics and Automation Letters (RA-L 2023)
Arxiv / Paper / Video / Github

We propose an active implicit object reconstruction method leveraging direct uncertainty evaluation from an implicit occupancy map and next-best-view optimization. It directly optimizes a virtual camera trajectory in the uncertainty field to maximize information gain for the current reconstruction.

Efficient Implicit Neural Reconstruction Using LiDAR
Dongyu Yan, Xiaoyang Lyu, Jieqi Shi, Yi Lin.
IEEE International Conference on Robotics and Automation (ICRA 2023)
Arxiv / Paper / Video / Github / Project Page

We propose an implicit reconstruction method that uses LiDAR scans as input, which is efficient, accurate, and applicable in various scenarios. This method can be applied to real-world scenes, even with sparse LiDAR scans and poorly aligned poses, as shown in the self-collected dataset we released.