Skip to content

KIT-MRT/masked-fusion-360

Repository files navigation

MaskedFusion360: Reconstruct LiDAR Data by Querying Camera Features arXiv

TL;DR: Self-supervised pre-training method to fuse LiDAR and camera features for self-driving applications.

Model architecture

Reconstruct LiDAR data by querying camera features. Spherical projections of LiDAR data are transformed into patches, afterwards, randomly selected patches are removed and a MAE encoder is applied to the unmasked patches. The encoder output tokens are fused with camera features via cross-attention. Finally, a MAE decoder reconstructs the spherical LiDAR projections.

Getting started

Coming soon...

Demo with data from our test vehicle Joy

Reconstruction demo

Acknowledgement

The masked autoencoder (He et al., 2022) and CrossViT (Chen et al., 2021) implementations are based on lucidrain's vit_pytorch library.

About

Self-supervised pre-training method to fuse LiDAR and camera features.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published