Song Jin

Song Jin (宋瑾)

Staff Algorithm Engineer

Alibaba Group — 大淘宝技术

I am a Staff Algorithm Engineer at Alibaba Group, where I lead the algorithm development for Jiazuo, Alibaba's cutting-edge AI platform for home decoration.

My journey began with a strong foundation in SLAM and 3D Vision during my B.S. and M.S. at Harbin Institute of Technology (HIT) and Hikvision Research Institute. With extensive expertise in Image Diffusion Models, my current research interests have evolved towards the frontier of Object-centric 3D Generation and Physics-driven World Models. I am passionate about bridging the gap between generative AI and physical laws to create intelligent systems that can perceive and simulate the 3D world with high fidelity.

News

2025.12 Started research on 3D component-wise generation.
2025.11 Two preprints released: SkeleGuide (skeleton-guided human image synthesis) and SA-IQA (spatial aesthetics image quality assessment).
2024 Promoted to Staff Algorithm Engineer at Alibaba Group.
2024 Paper HomeDiffusion accepted at AAAI 2024.
2023.3 Leading algorithm development for jiazuo.taobao.com, Alibaba's AI home decoration image generation platform.

Publications

HomeDiffusion: Zero-shot Object Customization with Multi-view Representation Learning for Indoor Scenes

Song Jin, et al.

AAAI 2024

Paper

SkeleGuide: Explicit Skeleton Reasoning for Context-Aware Human-in-Place Image Synthesis

Song Jin, et al.

Preprint, 2025

Paper

SA-IQA: Redefining Image Quality Assessment for Spatial Aesthetics with Multi-Dimensional Rewards

Song Jin, et al.

Preprint, 2025

Paper

3D Room Layout Reconstruction from a Single RGB Image

Song Jin, et al.

2022

Paper

Projects

家作 — Home Decoration Image Generation

Algorithm Lead, 2023.3–Present — jiazuo.taobao.com

End-to-end AI image generation platform for home decoration scenes, covering studio photography (棚拍), AI model generation (模特), and suite-level commercialization (套图). Model evolution: SD+ControlNet → SDXL → Flux.

放我家 — Virtual Home Staging

2022.3–2024

AI-powered indoor scene editing pipeline integrating object removal (early adoption of SD inpainting), single-image 3D room layout reconstruction, and zero-shot object insertion with multi-view representation learning (HomeDiffusion).

户型扫描 — Floor Plan Scanning & Reconstruction

2021.5–2022.3

Corner-based floor plan scanning and LiDAR-based 3D reconstruction for interior spaces.

Video

Stereo Camera Self-Calibration

2020 — HikVision

On-chip stereo self-calibration without feature extraction. Achieved <0.5px epipolar alignment error with ~30s processing on HiSi3559.

Amphibious Robot VIO

2019 — HIT

Visual-inertial localization for amphibious robots in field environments with adaptive viewing angle and exposure control.

Code Video PDF

MAV SLAM

2018 — HIT

Autonomous state estimation and mapping for MAVs using onboard stereo cameras with EKF-based pose estimation and real-time 3D perception.

Code PDF

Experience

2024 — Present

Staff Algorithm Engineer

Alibaba Group

大淘宝技术

2021.2 — 2024

Senior Algorithm Engineer

Alibaba Group

大淘宝技术

2020.4 — 2021.2

Algorithm Engineer

HikVision Research Institute

2017.9 — 2020.1

M.S. in Computer Science

Harbin Institute of Technology

2013.9 — 2017.9

B.S. in Computer Science

Harbin Institute of Technology

Blog

Coming soon.