A fair comparison should be based on the same protocol--comments on “trainable convolution filters and their
IEEE Transactions on Pattern Analysis and Machine Intelligence
|October 9, 2013
Summary
This study re-evaluates the Volterra kernel classifier (Volterrafaces) for face recognition. Unfair comparison settings were identified, showing state-of-the-art performance on only one of three tested databases.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- The Volterra kernel classifier, termed Volterrafaces for face recognition, has been proposed as an image classification approach.
- Performance evaluations were conducted using experiments on face recognition databases.
Purpose of the Study:
- To critically assess the reported performance of the Volterra kernel classifier (Volterrafaces) in face recognition.
- To re-evaluate the method's performance under standard experimental protocols and compare it against the state-of-the-art.
Main Methods:
- The study involved analyzing the experimental settings used in a previous paper on Volterrafaces.
- New performance results were generated using standard protocols on three face recognition datasets.
- Comparisons were made with existing state-of-the-art methods on these datasets.
Main Results:
- The original comparisons with state-of-the-art methods were found to be based on unfair settings.
- Under standard protocols, Volterrafaces achieved state-of-the-art performance on only one of the three evaluated face recognition databases.
Conclusions:
- The reported superior performance of Volterrafaces in the original study is questionable due to non-standard comparison settings.
- The method's effectiveness is dataset-dependent, achieving top-tier results on only a subset of benchmarks.
