UniPR-3D / README.md
nielsr's picture
nielsr HF Staff
Add model card and metadata
047bfab verified
|
raw
history blame
1.69 kB
metadata
license: mit
pipeline_tag: image-feature-extraction

UniPR-3D: Towards Universal Visual Place Recognition with Visual Geometry Grounded Transformer

UniPR-3D is a universal visual place recognition (VPR) framework that effectively integrates information from multiple views. It supports both frame-to-frame and sequence-to-sequence matching by leveraging 3D and 2D tokens with tailored aggregation strategies.

Description

UniPR-3D builds on a Visual Geometry Grounded Transformer (VGGT) backbone capable of encoding multi-view 3D representations. To construct its descriptor, the model jointly leverages 3D tokens and intermediate 2D tokens, using dedicated aggregation modules to capture fine-grained texture cues while reasoning across viewpoints. To further enhance generalization, it incorporates both single- and multi-frame aggregation schemes along with a variable-length sequence retrieval strategy. It achieves state-of-the-art performance on several benchmarks, including MSLS, Pittsburgh, NordLand, and SPED.

Citation

If you find our paper and code useful, please cite us:

@inproceedings{deng2026_unipr3d,
  title     = {UniPR-3D: Towards Universal Visual Place Recognition with 3D Visual Geometry Grounded Transformer},
  author    = {Tianchen Deng and Xun Chen and Ziming Li and Hongming Shen and Danwei Wang and Javier Civera and Hesheng Wang},
  booktitle = {Arxiv},
  year      = {2026},
}