Haoge Deng

I am now a visiting student at the NLPR, Institute of Automation, Chinese Academy of Sciences (CASIA), supervised by Prof Zhaoxiang Zhang. I am also a final-year MPhil. student at BUPT in China, supervised by Prof. Yonggang Qi. Prior to that, I obtained my Bachelor's degree in Electronics Information Science and Technology at BUPT in 2022.

My research interests revolve around computer visionn and generative models , with a particular focus on image generation , video generation , and 3D content creation .

Email  /  Scholar  /  Github

profile photo

Research

Representative papers are highlighted. * indicates equal contribution, # indicates corresponding author.

Autoregressive Video Generation without Vector Quantization
Haoge Deng*, Ting Pan*, Haiwen Diao*, Zhengxiong Luo*, Yufeng Cui, Huchuan Lu, Shiguang Shan, Yonggang Qi#, Xinlong Wang#
International Conference on Learning Representations (ICLR, TH-CPL A), 2025
[arxiv] | [project page] | [code] | [openreview] | [post] | [hugging face 🤗 daily papers]

NOVA is a non-quantized autoregressive model that enables efficient video generation by reformulating the video creation as frame-by-frame and set-by-set predictions.

You See it, You Got it: Learning 3D Creation on Pose-Free Videos at Scale
Baorui Ma*, Huachen Gao*, Haoge Deng*, Zhengxiong Luo, Tiejun Huang, Lulu Tang# Xinlong Wang#
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR, CCF-A), 2025
[arxiv] | [project page] | [code] | [dataset] | [post] | [hugging face 🤗 daily papers]

See3D is a scalable visual-conditional MVD model for open-world 3D creation, which can be trained on web-scale video collections without camera pose annotations.

GeoDream: Disentangling 2D and Geometric Priors for High-Fidelity and Consistent 3D Generation
Baorui Ma*, Haoge Deng*, Junsheng Zhou , Yu-Shen Liu, Tiejun Huang, Xinlong Wang#
arXiv, 2023
[arXiv] | [project page] | [code]

GeoDream is a 3D generation method that integrates explicit generalized 3D priors with 2D diffusion priors to enhance the capability of obtaining unambiguous 3D consistent geometric structures without sacrificing diversity or fidelity.

SketchKnitter: Vectorized Sketch Generation with Diffusion Models
Qiang Wang, Haoge Deng, Yonggang Qi#, Da Li, Yi-Zhe Song,
International Conference on Learning Representations (ICLR, TH-CPL A), 2023 (Spotlight, ~5% acceptance rate)
[paper] | [openreview] | [code]

SketchKnitter is a method that achieves vectorized sketch generation by reversing the stroke deformation process using a diffusion model learned from real sketches, enabling the creation of higher quality, visually appealing sketches with fewer sampling steps.

Education


BUPT
Beijing University of Posts and Telecommunications
Master of Science in Artificial Intelligence
Under the supervision of Prof. Yonggang Qi
Sept. 2022 ~ July. 2025 (Expected)
BUPT
Beijing University of Posts and Telecommunications
Bachelor of Engineering in Electronics Engineering
Outstanding Graduate of Beijing Province
Sept. 2018 ~ July. 2022

Experiences


baailogo

Beijing Academy of Artificial Intelligence
Research Intern at BAAI-Vision
Research on Video generation
Advised by Dr. Zhengxiong Luo and Dr. Xinlong Wang
Jun. 2024 ~ Now

baailogo

Beijing Academy of Artificial Intelligence
Research Intern at BAAI-Vision
Research on 3D content generation
Advised by Dr. Baorui Ma and Dr. Xinlong Wang
Jun. 2023 ~ Jun. 2024

MeiTuan

MeiTuan
Research Intern at Meituan-AutoML
Research on Lidar lidar range generation
Advised by Dr. Zhi Tian and Dr. Xiangxiang Chu
Jan. 2023 ~ Jun. 2023

Academic Services

  • Conference Reviewer: ICASSP'{24,25}, ICCV'{25}
  • Workshop Reviewer: ICLR'{25}-DeLTa.

Website Template