Systematic evaluation of CNN advances on the ImageNet

TL;DR

This paper systematically evaluates recent CNN architectural and training advances on ImageNet, analyzing their individual and combined effects to identify optimal configurations efficiently.

Contribution

It provides a comprehensive, systematic comparison of CNN design choices and training methods, highlighting their independent contributions and practical implications.

Findings

Individual modifications improve performance, with combined gains showing near-independence.

128x128 images suffice for qualitative analysis, enabling faster experimentation.

The study offers guidelines for optimizing CNN architectures on ImageNet.

Abstract

The paper systematically studies the impact of a range of recent advances in CNN architectures and learning methods on the object categorization (ILSVRC) problem. The evalution tests the influence of the following choices of the architecture: non-linearity (ReLU, ELU, maxout, compatibility with batch normalization), pooling variants (stochastic, max, average, mixed), network width, classifier design (convolutional, fully-connected, SPP), image pre-processing, and of learning parameters: learning rate, batch size, cleanliness of the data, etc. The performance gains of the proposed modifications are first tested individually and then in combination. The sum of individual gains is bigger than the observed improvement when all modifications are introduced, but the "deficit" is small suggesting independence of their benefits. We show that the use of 128x128 pixel images is sufficient to make qualitative conclusions about optimal network structure that hold for the full size Caffe and VGG nets. The results are obtained an order of magnitude faster than with the standard 224 pixel images.