A Caltech Library Service

Anchor Loss: Modulating Loss Scale Based on Prediction Difficulty

Ryou, Serim and Jeong, Seong-Gyun and Perona, Pietro (2019) Anchor Loss: Modulating Loss Scale Based on Prediction Difficulty. In: 2019 IEEE/CVF International Conference on Computer Vision (ICCV). IEEE , Piscataway, NJ, pp. 5991-6000. ISBN 9781728148038.

[img] PDF - Submitted Version
See Usage Policy.


Use this Persistent URL to link to this item:


We propose a novel loss function that dynamically re-scales the cross entropy based on prediction difficulty regarding a sample. Deep neural network architectures in image classification tasks struggle to disambiguate visually similar objects. Likewise, in human pose estimation symmetric body parts often confuse the network with assigning indiscriminative scores to them. This is due to the output prediction, in which only the highest confidence label is selected without taking into consideration a measure of uncertainty. In this work, we define the prediction difficulty as a relative property coming from the confidence score gap between positive and negative labels. More precisely, the proposed loss function penalizes the network to avoid the score of a false prediction being significant. To demonstrate the efficacy of our loss function, we evaluate it on two different domains: image classification and human pose estimation. We find improvements in both applications by achieving higher accuracy compared to the baseline methods.

Item Type:Book Section
Related URLs:
URLURL TypeDescription Paper
Perona, Pietro0000-0002-7583-5809
Additional Information:© 2019 IEEE. We would like to thank Joseph Marino and Matteo Ruggero Ronchi for their valuable comments. This work was supported by funding from Disney Research.
Funding AgencyGrant Number
Disney ResearchUNSPECIFIED
Record Number:CaltechAUTHORS:20200306-125849462
Persistent URL:
Official Citation:S. Ryou, S. Jeong and P. Perona, "Anchor Loss: Modulating Loss Scale Based on Prediction Difficulty," 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea (South), 2019, pp. 5991-6000. doi: 10.1109/ICCV.2019.00609
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:101742
Deposited By: Tony Diaz
Deposited On:06 Mar 2020 21:11
Last Modified:16 Nov 2021 18:05

Repository Staff Only: item control page