Improving Map-Free Localization with Depth Supervision
Résumé
We propose an enhancement to the MicKey architecture, a state-of-the-art method for map-free localization, by integrating the Depth Anything network [10] to improve depth estimation. Our approach uses Depth Anything-generated ground truth depth maps during MicKey's training, resulting in more accurate depth predictions, especially in challenging regions like depth discontinuities. Experimental results show that our method maintains competitive performance across key metrics while requiring minimal supervision, thereby improving the versatility and accessibility of the MicKey framework for diverse datasets.
Origine | Fichiers produits par l'(les) auteur(s) |
---|