Video Depth Estimation by Fusing Flow-to-Depth Proposals
This enables devices and robots with single cameras to perceive 3D depth, though it is an incremental improvement combining existing components like optical flow and fusion networks.
The paper tackles video depth estimation from monocular videos by introducing a differentiable flow-to-depth layer that generates depth proposals through epipolar geometry optimization, along with camera pose refinement and depth fusion modules. The approach outperforms state-of-the-art methods on three public datasets and shows good cross-dataset generalization, such as training on KITTI and performing well on Waymo.
Depth from a monocular video can enable billions of devices and robots with a single camera to see the world in 3D. In this paper, we present an approach with a differentiable flow-to-depth layer for video depth estimation. The model consists of a flow-to-depth layer, a camera pose refinement module, and a depth fusion network. Given optical flow and camera pose, our flow-to-depth layer generates depth proposals and the corresponding confidence maps by explicitly solving an epipolar geometry optimization problem. Our flow-to-depth layer is differentiable, and thus we can refine camera poses by maximizing the aggregated confidence in the camera pose refinement module. Our depth fusion network can utilize depth proposals and their confidence maps inferred from different adjacent frames to produce the final depth map. Furthermore, the depth fusion network can additionally take the depth proposals generated by other methods to improve the results further. The experiments on three public datasets show that our approach outperforms state-of-the-art depth estimation methods, and has reasonable cross dataset generalization capability: our model trained on KITTI still performs well on the unseen Waymo dataset.