Dense CNN Learning with Equivalent Mappings

TL;DR

This paper introduces eConv and ePool layers that enable dense, high-accuracy predictions in CNNs while maintaining equivalence to the baseline model, improving transferability and speed.

Contribution

The paper proposes novel eConv and ePool layers that produce dense, equivalent predictions, facilitating parameter transfer and improving accuracy across multiple vision tasks.

Findings

eConv and ePool achieve higher accuracy than baseline CNNs.

The methods enable parameter transfer between dense and original models.

The approach improves both speed and accuracy in various tasks.

Abstract

Large receptive field and dense prediction are both important for achieving high accuracy in pixel labeling tasks such as semantic segmentation. These two properties, however, contradict with each other. A pooling layer (with stride 2) quadruples the receptive field size but reduces the number of predictions to 25\%. Some existing methods lead to dense predictions using computations that are not equivalent to the original model. In this paper, we propose the equivalent convolution (eConv) and equivalent pooling (ePool) layers, leading to predictions that are both dense and equivalent to the baseline CNN model. Dense prediction models learned using eConv and ePool can transfer the baseline CNN's parameters as a starting point, and can inverse transfer the learned parameters in a dense model back to the original one, which has both fast testing speed and high accuracy. The proposed eConv and ePool layers have achieved higher accuracy than baseline CNN in various tasks, including semantic segmentation, object localization, image categorization and apparent age estimation, not only in those tasks requiring dense pixel labeling.