Towards CNN map representation and compression for camera relocalisation
This work addresses map compression for camera relocalization in robotics or AR/VR, but it appears incremental as it builds on existing CNN methods.
The paper tackles camera relocalization by using Convolutional Neural Networks (CNNs) for map representation and compression, achieving improved accuracy by increasing training trajectories while keeping CNN size constant.
This paper presents a study on the use of Convolutional Neural Networks for camera relocalisation and its application to map compression. We follow state of the art visual relocalisation results and evaluate the response to different data inputs. We use a CNN map representation and introduce the notion of map compression under this paradigm by using smaller CNN architectures without sacrificing relocalisation performance. We evaluate this approach in a series of publicly available datasets over a number of CNN architectures with different sizes, both in complexity and number of layers. This formulation allows us to improve relocalisation accuracy by increasing the number of training trajectories while maintaining a constant-size CNN.