Sequential Feature Filtering Classifier

TL;DR

The paper introduces Sequential Feature Filtering Classifier (FFC), a simple method that improves CNN performance across various tasks by filtering features through sequential normalization and activation, with minimal computational overhead.

Contribution

FFC is a novel classifier that sequentially filters features in CNNs, enhancing performance across multiple vision tasks with negligible additional computation.

Findings

FFC significantly improves accuracy on ImageNet-1K classification.

FFC enhances detection and segmentation results on MS COCO and Cityscapes.

FFC further boosts performance when combined with attention and augmentation techniques.

Abstract

We propose Sequential Feature Filtering Classifier (FFC), a simple but effective classifier for convolutional neural networks (CNNs). With sequential LayerNorm and ReLU, FFC zeroes out low-activation units and preserves high-activation units. The sequential feature filtering process generates multiple features, which are fed into a shared classifier for multiple outputs. FFC can be applied to any CNNs with a classifier, and significantly improves performances with negligible overhead. We extensively validate the efficacy of FFC on various tasks: ImageNet-1K classification, MS COCO detection, Cityscapes segmentation, and HMDB51 action recognition. Moreover, we empirically show that FFC can further improve performances upon other techniques, including attention modules and augmentation techniques. The code and models will be publicly available.