Multi-task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics

TLDR

Multitask learning with regression and classification objectives benefits many deep learning applications, but manually tuning task weights is difficult and expensive, limiting its practical use. We propose to weight each task’s loss by its homoscedastic uncertainty, enabling simultaneous learning of depth regression, semantic and instance segmentation from a single monocular image across tasks with different units or scales. The model learns per‑pixel depth, semantic, and instance segmentation from a monocular input image. It outperforms separate single‑task models trained individually on each task.

Abstract

Numerous deep learning applications benefit from multitask learning with multiple regression and classification objectives. In this paper we make the observation that the performance of such systems is strongly dependent on the relative weighting between each task's loss. Tuning these weights by hand is a difficult and expensive process, making multi-task learning prohibitive in practice. We propose a principled approach to multi-task deep learning which weighs multiple loss functions by considering the homoscedastic uncertainty of each task. This allows us to simultaneously learn various quantities with different units or scales in both classification and regression settings. We demonstrate our model learning per-pixel depth regression, semantic and instance segmentation from a monocular input image. Perhaps surprisingly, we show our model can learn multi-task weightings and outperform separate models trained individually on each task.