A General and Adaptive Robust Loss Function

By Jonathan T. Barron. Computer Vision and Pattern Recognition, 2017

763 citations recorded by the source corpus.

We present a generalization of the Cauchy/Lorentzian, Geman-McClure, Welsch/Leclerc, generalized Charbonnier, Charbonnier/pseudo-Huber/L1-L2, and L2 loss functions. By introducing robustness as a continuous parameter, our loss function allows algorithms built around robust loss minimization to be generalized, which improves performance on basic vision tasks such as registration and clustering. Interpreting our loss as the negative log of a univariate density yields a general probability distribution that includes normal and Cauchy distributions as special cases. This probabilistic interpretation enables the training of neural networks in which the robustness of the loss automatically adapts itself during training, which improves performance on learning-based tasks such as generative image synthesis and unsupervised monocular depth estimation, without requiring any manual parameter tuning.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/9b67eec05ed92afed099c73695989c64aa49e3cd

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

Citations in the corpus

Zip-NeRF: Anti-Aliased Grid-Based Neural Radiance Fields (2023, methodology): “We therefore use a more challenging evaluation procedure, similar to the multiscale Blender dataset used by mip-NeRF [3]: we turn each image into a set of four images that have been bicubically downsampled by scale factors [1, 2, 4, 8] where the downsampled images act as a proxy for additional train”
Human Pose Regression with Residual Log-likelihood Estimation (2021, background): “Barron [2] presents a generalization of common loss functions, which automatically adapts itself during training.”
Nerfies: Deformable Neural Radiance Fields (2020, methodology): “where ρ(·) is the Geman-McClure robust error function [19] implemented as per Barron [3] with hyperparameter c = 0.”
HITNet: Hierarchical Iterative Tile Refinement Network for Real-time Stereo Matching (2020, methodology): “We use the general robust loss function ρ(·) from [2] which resembles a smooth 1 loss, i.”
AIM 2020 Challenge on Real Image Super-Resolution: Methods and Results (2020, methodology): “p 29.89 SM + F p p p 30.08 SE + F p p p 30.25 For training, L 1, multi-scale SSIM and VGG loss functions are considered in the model, dened as follows L f = L 1(^y;y) + L MS-SSIM(y^;y) + L VGG(y^;y) (2) L VGG uses the features of conv2 layer after ReLU in the pre-trained VGG-16 network. Three RRGs a”
RoMa: Robust Dense Feature Matching (2023, positive): “We next discuss the fine loss L fine . of We model the output the refinement at scale i as a generalized Charbonnier [3] (with α = 0 .”
Simple and Effective VAE Training with Calibrated Decoders (2020, neutral): “(Takahashi et al., 2018; Barron, 2019) proposed more expressive distributions.”