Excavating the Potential Capacity of Self-Supervised Monocular Depth Estimation

TL;DR

This paper demonstrates that the potential of self-supervised monocular depth estimation can be fully realized without extra annotation costs by introducing novel data augmentation, self-distillation, and network design techniques, leading to state-of-the-art results.

Contribution

The paper introduces a new data augmentation method, a self-distillation loss with a novel post-processing technique, and a full-scale network architecture to enhance self-supervised depth estimation.

Findings

Significant performance improvements over baseline models.

EPCDepth surpasses previous state-of-the-art methods.

Achieves these results with less computational overhead.

Abstract

Self-supervised methods play an increasingly important role in monocular depth estimation due to their great potential and low annotation cost. To close the gap with supervised methods, recent works take advantage of extra constraints, e.g., semantic segmentation. However, these methods will inevitably increase the burden on the model. In this paper, we show theoretical and empirical evidence that the potential capacity of self-supervised monocular depth estimation can be excavated without increasing this cost. In particular, we propose (1) a novel data augmentation approach called data grafting, which forces the model to explore more cues to infer depth besides the vertical image position, (2) an exploratory self-distillation loss, which is supervised by the self-distillation label generated by our new post-processing method - selective post-processing, and (3) the full-scale network, designed to endow the encoder with the specialization of depth estimation task and enhance the representational power of the model. Extensive experiments show that our contributions can bring significant performance improvement to the baseline with even less computational overhead, and our model, named EPCDepth, surpasses the previous state-of-the-art methods even those supervised by additional constraints.