DeMoN: Depth and Motion Network for Learning Monocular Stereo
This work addresses the challenge of monocular stereo estimation for computer vision applications, offering a novel learning-based approach that is incremental in combining depth and motion estimation.
The paper tackles the problem of estimating depth and camera motion from unconstrained image pairs by formulating structure from motion as a learning problem, resulting in more accurate and robust outcomes compared to traditional methods, with improved generalization to unseen structures.
In this paper we formulate structure from motion as a learning problem. We train a convolutional network end-to-end to compute depth and camera motion from successive, unconstrained image pairs. The architecture is composed of multiple stacked encoder-decoder networks, the core part being an iterative network that is able to improve its own predictions. The network estimates not only depth and motion, but additionally surface normals, optical flow between the images and confidence of the matching. A crucial component of the approach is a training loss based on spatial relative differences. Compared to traditional two-frame structure from motion methods, results are more accurate and more robust. In contrast to the popular depth-from-single-image networks, DeMoN learns the concept of matching and, thus, better generalizes to structures not seen during training.