Recently, the China Computer Federation (CCF) announced the acceptance results for the top-tier international conference on computer vision and pattern recognition, CVPR, which is an A-class conference in the field of computer vision. The research work on large-model-based 3D visual localization, titled "Chain of Semantics Programming in 3D Gaussian Splatting Representation for 3D Vision Grounding," was accepted for presentation. The research was conducted by Shi Jiaxin and Xiang Mingyue, undergraduate students from the 2022 class of the School of Electronic Information Engineering, Huang Yixuan, a 2023 undergraduate student, and Sun Hao, an undergraduate from the 2022 class of the School of Transportation.

CVPR, the IEEE International Conference on Computer Vision and Pattern Recognition, is one of the most prestigious conferences in the field of computer vision, recommended by CCF as an A-class international academic conference. According to the official statistics of the conference, CVPR 2025 received a total of 13,008 submissions, with 2,878 accepted, resulting in an acceptance rate of only 22.1%.

To address the challenges of acquiring fine-grained semantic information and reasoning complex spatial relationships in 3D visual localization tasks, this research proposed a zero-shot neural symbolic model. The model decomposes complex spatial relationships through semantic chain programming and utilizes 3DGS representation to provide fine-grained semantics. The study was evaluated on two public datasets, with experimental results showing that the model achieved 60.8% accuracy on the Nr3D dataset, approaching the performance of the latest supervised models, and 91.4% accuracy on the Sr3D dataset, outperforming the latest supervised models.

In this research, our university is the only contributing institution, with Prof. Weng Zhi from the School of Electronic Information Engineering as the corresponding author. This work was supported by the National Innovation and Entrepreneurship Training Program for College Students (New Engineering Key Areas Support Project, Project No. 202410126042, Guided by Prof. Weng Zhi).