FlashMix

Fig. 1: Comparision of LiDAR pose regression-based framework (top) with our fast map-free LiDAR localization system.

Contributions

Map-free LiDAR localization framework combining a pre-trained point encoder and a scene-specific pose regressor, with feature buffer enabled rapid training.
MLP-Mixer based descriptor aggregator to fuse global point relationships by feature mixing for adapting to scene-specific geometries.
Contrastive and metric learning loss based regularization to enhance global descriptor retrieval performance and stable convergence while maintaining fast training times.
Extensive experiments in outdoor and indoor environments, demonstrating rapid training and adaptation with competitive performance compared to existing map-free localization methods.

Video

Results

The relocalization rates of the best-performing models as a function of training time are plotted below.

Fig. 2: Analysis of relocalization rate as a function of train time

Visualizations

Fig. 3a: Oxford-Radar

Fig. 3b: Mulran DCC

Fig. 3c: vReLoc

Fig. 3: Visualization of different methods on test trajectories from Oxford-Radar, DCC, and vReLoC datasets. Trajectory visualization: The ground truth and estimated positions are shown in dark blue and red dots, respectively. The star shows the starting position.

Citation

@inproceedings{goswami2025flashmix,
  title={Flashmix: Fast map-free LiDAR localization via feature mixing and contrastive-constrained accelerated training},
  author={Goswami, Raktim Gautam and Patel, Naman and Krishnamurthy, Prashanth and Khorrami, Farshad},
  booktitle={2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  pages={2011--2020},
  year={2025},
  organization={IEEE}
}