I’m currently a final-year PhD. Candidate from Hong Kong PolyU (AAE) and will be graduated within 2026. My PhD. Thesis is about Promoting Visual Perception under Challenging Underwater Environment: From 2D to 3D Applications. I am supervised by Prof. Chih-yung Wen and co-supervised by Prof. Bing Wang. Before I pursuit my doctoral degree, I received a master degree with distinction also from PolyU supervised by Prof. Boyang Li in 2023, and a bachelor degree from Nanjing University of Aeronautics and Astronautics (NUAA) supervised by Prof. Meng Yu in 2019. I also worked as a research assistant at Dept. of Automation, Tsinghua University from 2019 to 2021, supervised by Prof. Zhang Chen.

I am actively seeking research-oriented positions in the industry, with a focus on underwater visual intelligence — especially in areas related to 3D reconstruction and embodied AI. Please feel free to reach out if you have relevant opportunities.

Research Focus

My PhD. research consistently locates in enhancing the underwater images degraded by water medium (including color-bias, low-contrast, and blur effects that weaken the vision-based tasks) to support both 2D visual evaluation and downstream 3D reconstruction. Specifically, I’ve spent much effort to solve the issues of label noise and multi-view inconsistency that impair the model performance by physics-informed and learning-based strategies.

2D Evaluation-oriented Enhancement

• Underwater sequential images enhancement via diffusion and physics priors fusion

  This work focuses on:

  1. leveraging diffusion prior to achieve SOTA performance from only synthetic data;
  2. full physics-based synthesis for improving visual clarity.

• Fusing Transferred Priors and Physics-based Decomposition for Underwater Image Enhancement

  This work focuses on:

  1. transfer-learning from multiple data/model priors to avoid noisy label issue;
  2. Physics-based task decomposition to provide theoretical soundness.

3D Reconstruction-oriented Enhancement

• NVS-UIE: Diffusion UIE towards Novel View Synthesis via Physics-Aligned Frequency Fusion

  This work focuses on:

  1. frequency-fusion strategy to preserve the texture details;
  2. physics-controlled sequential image synthesis to improve the multi-view consistency.

• RQUL-UIE: Revitalizing Quality-Unstable Labels for Underwater Image Enhancement via In-Dataset Self-Supervision

  This work focuses on:

  1. label quality quantization to fully utilize the quality-unstable labels;
  2. a fourier-based texture refinement to for multi-view consistent texture preservation.

Selected Publications

• RQUL-UIE: Revitalizing Quality-Unstable Labels for Underwater Image Enhancement via In-Dataset Self-Supervision
  Haochen Hu, Yanrui Bin, Chih-yung Wen, Bing Wang* Prepared to submitted tot CVPR 2027.
  [Paper] [Code]

• Fusing Transferred Priors and Physics-based Decomposition for Underwater Image Enhancement
  Haochen Hu, Yanrui Bin, Zhengyan Zhang, Minchen Wei, Chih-yung Wen, Bing Wang*
  Information Fusion IF:15.5 (Early accepted).
  [Paper] [Code]

• Underwater sequential images enhancement via diffusion and physics priors fusion
  Haochen Hu, Yanrui Bin, Chih-yung Wen, Bing Wang*
  Information Fusion IF:15.5.
  [Paper]

• NVS-UIE: Diffusion UIE towards Novel View Synthesis via Physics-Aligned Frequency Fusion
  Haochen Hu, Yanze Lu, Yuanbo Xue, Zhengyan Zhang, Shu Yang, Chih-yung Wen, Bing Wang*
  Visual Intelligence (Early accepted).
  [Paper]