Increasing detection performance under stress thank to gradient regularisation.

Adrien Chan-Hon-TongINRIA 2019

Abstract

Even if current deep detectors are quite efficient, there is still a long road before dependable detection. First step on this road is performance stability under simple stresses like Gaussian noise, blur or adversarial noise. This article investigates deep detector performance stability under stress, and, precisely, effect of different kind of Lipschitz based regularity. As a result, gradient penalty not only improve stability under stress but also raw performances.
BibTeX
@inproceedings{chanhontong2019increasing,
  author = {Adrien Chan-Hon-Tong},
  title = {{Increasing detection performance under stress thank to gradient regularisation.}},
  booktitle = {INRIA},
  year = {2019},
}

Loading related content...

Figure