FasterDepth: a lightweight network for self-supervised monocular depth estimation

Self-supervised monocular depth estimation uses only one camera to get depth information. It does not need manually real depth information as training data, but is trained by the geometric information contained in the image itself. While many existing methods use heavy backbone networks for precision, designing lightweight models can reduce the computational and memory consumption, making them suitable for resource-constrained environments or embedded devices. In this work, we propose a lightweight network (FasterDepth) for self-supervised monocular depth estimation. Additionally, in order to merge the rich information of multi-stage of the network, this work raises a multi-stage feature fusion module. Experiments on the KITTI dataset show that our FasterDepth has higher precision and fewer parameters than Monodepth2.