Haruko386

🍀About me

Come from Taiyuan, Shanxi, China, currently pursuing a master’s degree in computer science at ShenYang Jianzhu University, supervised by Prof.Yuanshuai. For now, my primary research focuses on fine-tuning diffusion models for deep estimation tasks.

I used to participate in many ACM algorithm competitions(2022-2023), but have now retired. In my daily life, I enjoy reading and photography. You can view my photography collection in the Gallery. I’m also a Golang backend developer. Feel free to communicate with me about CV and Golang

📝 Publications

axXiv

ApDepth V2: Look at or look through the window

[~~Paper~~] [~~Model~~]

We introduce ApDepth V2, an enhanced single-step monocular depth estimation framework built upon stable-diffusion-xl-base-1.0. It introduces a continuous fine-tuning strategy with delayed decoder adaptation and Spatially Debiased Window Transport (SDWT), enabling misalignment-tolerant local geometry supervision while preserving depth reconstruction consistency.

arXiv

Mind the Sky: Resolving Sky-Depth Collapse with Validity-Guided Diffusion

[Paper(Ablation underway)] [Code] [~~Model~~]

We presents Celeste, a multi-step latent diffusion framework that addresses this gap through validity-guided completion. The method separates fully valid, fully invalid, and mixed latent cells, applies weak region-level anchoring and interior smoothness on invalid support, and adds a selective far-fieldanchor to reduce the ambiguity of distant regions.

IMAVIS 2026

ApDepth: Aiming for Precise Monocular Depth Estimation Based on Diffusion Models

[Paper] [Code] [Demo] [Model] [Bibtex]

This paper presents ApDepth, a novel single-step diffusion framework for monocular depth estimation that achieves fast inference while preserving fine-grained edge details through a tailored two-stage training strategy combined with novel frequency-domain and cosine similarity losses.

Please cite our paper if you find it useful:

@misc{wang2026apdepth,
  title  = {{ApDepth}: Aiming for Precise Monocular Depth Estimation Based on Diffusion Models},
  author = {Wang, Jiawei and Lei, Mingbo and Yao, Liang and Li, Jiayi and Lu, Tianyu and Chen, Yibo},
  year   = {2026},
  doi    = {10.2139/ssrn.7604741},
  url    = {https://doi.org/10.2139/ssrn.7604741},
  note   = {SSRN Preprint}
}

🤗Dataset

huggingface

Overwatch Depth Dataset

[Dataset]

The Overwatch Depth Dataset contains pixel-aligned RGB images and dense depth maps captured from diverse synthetic outdoor environments. Streets, buildings, vegetation, open spaces, and long-range sky regions provide varied geometry for training and evaluating monocular depth estimation models.

🎈ACM ＆ OI

ACM Algorithm Competition participant, primarily competing in online contests on platforms such as Leetcode, Atcoder, and Codeforces. Historical ratings on each platform are available here.

🏆Speedrun

I have participated in many speedrun competitions, mainly working on the Hollow Knight IL project, and have achieved top 30 worldwide rankings in each of them, even winning a WR in one of them and a fourth place in another.

WR on Hollow Knight Trial of the Warrior