Unsupervised Monocular Depth Learning with Integrated Intrinsics and Spatio-Temporal Constraints

TL;DR

This paper introduces an unsupervised monocular depth learning framework that predicts at-scale depth, egomotion, and camera intrinsics from monocular sequences, using spatial and temporal constraints to improve accuracy and reduce implementation complexity.

Contribution

It presents a single-network approach that estimates depth, pose, and intrinsics from monocular sequences with integrated geometric constraints, reducing training time and complexity.

Findings

Achieves state-of-the-art performance on KITTI dataset

Reduces training time compared to previous methods

Predicts at-scale depth and intrinsics without labeled data

Abstract

Monocular depth inference has gained tremendous attention from researchers in recent years and remains as a promising replacement for expensive time-of-flight sensors, but issues with scale acquisition and implementation overhead still plague these systems. To this end, this work presents an unsupervised learning framework that is able to predict at-scale depth maps and egomotion, in addition to camera intrinsics, from a sequence of monocular images via a single network. Our method incorporates both spatial and temporal geometric constraints to resolve depth and pose scale factors, which are enforced within the supervisory reconstruction loss functions at training time. Only unlabeled stereo sequences are required for training the weights of our single-network architecture, which reduces overall implementation overhead as compared to previous methods. Our results demonstrate strong performance when compared to the current state-of-the-art on multiple sequences of the KITTI driving dataset and can provide faster training times with its reduced network complexity.