Fixing the train-test resolution discrepancy: FixEfficientNet

TL;DR

This paper introduces FixEfficientNet, a modified version of EfficientNet that corrects train-test resolution discrepancies, leading to significant accuracy improvements on ImageNet and establishing new state-of-the-art results.

Contribution

The paper presents FixEfficientNet, a novel training procedure that addresses train-test resolution mismatch, significantly enhancing EfficientNet performance.

Findings

FixEfficientNet-B0 achieves 79.3% top-1 accuracy on ImageNet.

FixEfficientNet-L2 with weak supervision reaches 88.5% top-1 accuracy.

Improvements hold across different evaluation protocols and datasets.

Abstract

This paper provides an extensive analysis of the performance of the EfficientNet image classifiers with several recent training procedures, in particular one that corrects the discrepancy between train and test images. The resulting network, called FixEfficientNet, significantly outperforms the initial architecture with the same number of parameters. For instance, our FixEfficientNet-B0 trained without additional training data achieves 79.3% top-1 accuracy on ImageNet with 5.3M parameters. This is a +0.5% absolute improvement over the Noisy student EfficientNet-B0 trained with 300M unlabeled images. An EfficientNet-L2 pre-trained with weak supervision on 300M unlabeled images and further optimized with FixRes achieves 88.5% top-1 accuracy (top-5: 98.7%), which establishes the new state of the art for ImageNet with a single crop. These improvements are thoroughly evaluated with cleaner protocols than the one usually employed for Imagenet, and particular we show that our improvement remains in the experimental setting of ImageNet-v2, that is less prone to overfitting, and with ImageNet Real Labels. In both cases we also establish the new state of the art.