Scene Coordinate Regression with Angle-Based Reprojection Loss for Camera Relocalization

Xiaotian Li (1) , Juha Ylioinas (1) , Jakob Verbeek (2) , Juho Kannala (3)

1 Aalto University
2 Thoth - Apprentissage de modèles à partir de données massives
3 CMV - Center for Machine Vision Research

Image-based camera relocalization is an important problem in computer vision and robotics. Recent works utilize convolutional neural networks (CNNs) to regress for pixels in a query image their corresponding 3D world coordinates in the scene. The final pose is then solved via a RANSAC-based optimization scheme using the predicted coordinates. Usually, the CNN is trained with ground truth scene coordinates, but it has also been shown that the network can discover 3D scene geometry automatically by minimizing single-view reprojection loss. However, due to the deficiencies of reprojection loss, the network needs to be carefully initialized. In this paper, we present a new angle-based reprojection loss which resolves the issues of the original reprojection loss. With this new loss function, the network can be trained without careful initialization, and the system achieves more accurate results. The new loss also enables us to utilize available multi-view constraints, which further improve performance.

Mots clés: scene coordinate regression; deep neural networks; camera relocalization

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Xiaotian Li, Juha Ylioinas, Jakob Verbeek, Juho Kannala. Scene Coordinate Regression with Angle-Based Reprojection Loss for Camera Relocalization. ECCV 2018 - Workshop Geometry Meets Deep Learning, Sep 2018, Munich, Germany. pp.229-245, ⟨10.1007/978-3-030-11015-4_19⟩. ⟨hal-01867143v2⟩