H-Net: Unsupervised Attention-based Stereo Depth Estimation Leveraging Epipolar Geometry
This addresses the challenge of acquiring accurate ground truth data for depth estimation in computer vision, offering an incremental improvement over existing unsupervised methods.
The paper tackles the problem of unsupervised stereo depth estimation by introducing H-Net, a deep-learning framework that leverages epipolar geometry and a Siamese autoencoder to refine stereo matching, achieving state-of-the-art results on KITTI2015 and Cityscapes datasets and closing the gap with supervised methods.
Depth estimation from a stereo image pair has become one of the most explored applications in computer vision, with most of the previous methods relying on fully supervised learning settings. However, due to the difficulty in acquiring accurate and scalable ground truth data, the training of fully supervised methods is challenging. As an alternative, self-supervised methods are becoming more popular to mitigate this challenge. In this paper, we introduce the H-Net, a deep-learning framework for unsupervised stereo depth estimation that leverages epipolar geometry to refine stereo matching. For the first time, a Siamese autoencoder architecture is used for depth estimation which allows mutual information between the rectified stereo images to be extracted. To enforce the epipolar constraint, the mutual epipolar attention mechanism has been designed which gives more emphasis to correspondences of features which lie on the same epipolar line while learning mutual information between the input stereo pair. Stereo correspondences are further enhanced by incorporating semantic information to the proposed attention mechanism. More specifically, the optimal transport algorithm is used to suppress attention and eliminate outliers in areas not visible in both cameras. Extensive experiments on KITTI2015 and Cityscapes show that our method outperforms the state-ofthe-art unsupervised stereo depth estimation methods while closing the gap with the fully supervised approaches.