Multi-Loss Weighting with Coefficient of Variations
This addresses the computational burden for researchers and practitioners in machine learning and computer vision, but it is incremental as it builds on existing loss weighting methods.
The paper tackles the problem of computationally expensive grid search for setting loss weights in multi-loss optimization by proposing a weighting scheme based on the coefficient of variations, which adapts weights during training without extra optimization, showing empirical validity on depth estimation and semantic segmentation tasks.
Many interesting tasks in machine learning and computer vision are learned by optimising an objective function defined as a weighted linear combination of multiple losses. The final performance is sensitive to choosing the correct (relative) weights for these losses. Finding a good set of weights is often done by adopting them into the set of hyper-parameters, which are set using an extensive grid search. This is computationally expensive. In this paper, we propose a weighting scheme based on the coefficient of variations and set the weights based on properties observed while training the model. The proposed method incorporates a measure of uncertainty to balance the losses, and as a result the loss weights evolve during training without requiring another (learning based) optimisation. In contrast to many loss weighting methods in literature, we focus on single-task multi-loss problems, such as monocular depth estimation and semantic segmentation, and show that multi-task approaches for loss weighting do not work on those single-tasks. The validity of the approach is shown empirically for depth estimation and semantic segmentation on multiple datasets.