DenoiseRep: Denoising Model for Representation Learning

TL;DR

DenoiseRep introduces a novel denoising model that enhances feature discrimination in representation learning by jointly extracting and denoising features, improving performance across various vision tasks and architectures.

Contribution

The paper proposes DenoiseRep, a unified framework that combines feature extraction and denoising, with a theoretical proof of its equivalence and computation-free denoising, applicable to multiple architectures.

Findings

Impressive improvements on vision tasks like re-identification, classification, detection, and segmentation.

Effective on CNN and Transformer architectures, including ResNet, ViT, Swin, and Vmamda.

Demonstrates stability and robustness across diverse datasets and tasks.

Abstract

The denoising model has been proven a powerful generative model but has little exploration of discriminative tasks. Representation learning is important in discriminative tasks, which is defined as "learning representations (or features) of the data that make it easier to extract useful information when building classifiers or other predictors". In this paper, we propose a novel Denoising Model for Representation Learning (DenoiseRep) to improve feature discrimination with joint feature extraction and denoising. DenoiseRep views each embedding layer in a backbone as a denoising layer, processing the cascaded embedding layers as if we are recursively denoise features step-by-step. This unifies the frameworks of feature extraction and denoising, where the former progressively embeds features from low-level to high-level, and the latter recursively denoises features step-by-step. After that, DenoiseRep fuses the parameters of feature extraction and denoising layers, and theoretically demonstrates its equivalence before and after the fusion, thus making feature denoising computation-free. DenoiseRep is a label-free algorithm that incrementally improves features but also complementary to the label if available. Experimental results on various discriminative vision tasks, including re-identification (Market-1501, DukeMTMC-reID, MSMT17, CUHK-03, vehicleID), image classification (ImageNet, UB200, Oxford-Pet, Flowers), object detection (COCO), image segmentation (ADE20K) show stability and impressive improvements. We also validate its effectiveness on the CNN (ResNet) and Transformer (ViT, Swin, Vmamda) architectures.