README.md

January 14, 2026 · View on GitHub

📖 TSGaussian: Semantic and Depth-Guided Target-Specific Gaussian Splatting from Sparse Views

🔥 2D detection, semantic segmentation, and 3D reconstruction of specific target objects;⭐ Separation of reconstructed 3D objects based on their semantic identities, enabling distinct handling of different semantic components; 🔥 High-quality 3D reconstruction and rendering in sparse-view scenarios without compromising reconstruction quality.

Paper(TSGaussian)

Liang Zhao, Zehan Bao, Yi Xie, Hong Chen, Yaohui Chen, Weifu Li

Huazhong Agricultural University

🚩 Updates

☑ We develop an algorithm that extends high-performing 2D scene understanding techniques to the 3D domain in sparse views.

☑ The code and data will be released after the paper's acceptance. Please stay tuned.

☐ Process your own data

Table of Contents

alt text

Consistent Targeted Semantic Segmentation

alt text

Semantic Constraints for 2D-to-3D

alt text

Multi-Scale Depth Regularization

alt text

Run the Demo

You can download the demo data from Google Drive.

After downloading, extract the data into the data/ directory.

bash script/train.sh plant1 1 4,5,8

Contact

If you have any question or collaboration needs, please email liweifu@mail.hzau.edu.cn.

Acknowledgement

This study is based on gaussian-grouping as well as DNGaussian. We appreciate their great codes.