Asymmetric Loss For Multi-Label Classification (2020-09-29T00:00:00.000000Z)

TL;DR

This paper introduces a novel asymmetric loss ("ASL"), which enables to dynamically down-weights and hard-thresholds easy negative samples, while also discarding possibly mislabeled samples and demonstrating ASL applicability for other tasks, such as single-label classification and object detection.

Abstract

In a typical multi-label setting, a picture contains on average few positive labels, and many negative ones. This positive-negative imbalance dominates the optimization process, and can lead to under-emphasizing gradients from positive labels during training, resulting in poor accuracy. In this paper, we introduce a novel asymmetric loss ("ASL"), which operates differently on positive and negative samples. The loss enables to dynamically down-weights and hard-thresholds easy negative samples, while also discarding possibly mislabeled samples. We demonstrate how ASL can balance the probabilities of different samples, and how this balancing is translated to better mAP scores. With ASL, we reach state-of-the-art results on multiple popular multi-label datasets: MS-COCO, Pascal-VOC, NUS-WIDE and Open Images. We also demonstrate ASL applicability for other tasks, such as single-label classification and object detection. ASL is effective, easy to implement, and does not increase the training time or complexity. Implementation is available at: https://github.com/Alibaba-MIIL/ASL.

Authors

Lihi Zelnik-Manor

3 papers

M. Protter

2 papers

Emanuel Ben Baruch

2 papers

TL;DR

Abstract

Authors

References53 items

2021 IEEE International Conference on Multimedia and Expo (ICME)

ImageNet-21K Pretraining for the Masses

Re-labeling ImageNet: from Single to Multi-Labels, from Global to Localized Labels

Attention-Driven Dynamic Graph Convolutional Network for Multi-label Image Recognition

Distribution-Balanced Loss for Multi-Label Classification in Long-Tailed Datasets

Multi-Label Image Recognition with Multi-Class Attentional Regions

TResNet: High Performance GPU-Dedicated Architecture

Cross-Modality Attention with Semantic Graph Embedding for Multi-Label Classification

Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection

Multi-Label Classification with Label Graph Superimposing

Imbalance Problems in Object Detection: A Review

Once for All: Train One Network and Specialize it for Efficient Deployment

Learning Semantic-Specific Graph Representation for Multi-Label Image Recognition

Multi-Label Image Recognition with Joint Class-Aware Map Disentangling and Label Correlation Embedding

MMDetection: Open MMLab Detection Toolbox and Benchmark

AutoAugment: Learning Augmentation Strategies From Data

Time Matters in Regularizing Deep Networks: Weight Decay and Data Augmentation Affect Early Learning Dynamics, Matter Little Near Convergence

EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks

Searching for MobileNetV3

Multi-Label Image Recognition With Graph Convolutional Networks

FCOS: Fully Convolutional One-Stage Object Detection

Learning a Deep ConvNet for Multi-Label Classification With Partial Labels

Class-Balanced Loss Based on Effective Number of Samples

A Baseline for Multi-Label Image Classification Using an Ensemble of Deep Convolutional Neural Networks

The Open Images Dataset V4

Multi-Label Image Classification via Knowledge Distillation from Weakly-Supervised Detection

A disciplined approach to neural network hyper-parameters: Part 1 - learning rate, batch size, momentum, and weight decay

Decoupled Weight Decay Regularization

Multi-label Image Recognition by Recurrently Discovering Attentional Regions

Improved Regularization of Convolutional Neural Networks with Cutout

Focal Loss for Dense Object Detection

Deep Extreme Multi-label Learning

Learning Spatial Regularization with Image-Level Supervisions for Multi-label Image Classification

Learning Deep Representation for Imbalanced Classification

CNN-RNN: A Unified Framework for Multi-label Image Classification

Deep Residual Learning for Image Recognition

Exploit Bounding Box Annotations for Multi-Label Object Recognition

The Pascal Visual Object Classes Challenge: A Retrospective

Microsoft COCO: Common Objects in Context

NUS-WIDE: a real-world web image database from National University of Singapore

Classification with a Reject Option using a Hinge Loss

Bi-Modal Learning With Channel-Wise Attention for Multi-Label Image Classification

multi-label image classiﬁcation

Maximizing Subset Accuracy with Recurrent Neural Networks in Multi-label Classification

ON COMPLETENESS

The PASCAL Visual Object Classes Challenge

The Pascal Visual Object Classes Challenge 2006 ( VOC 2006 ) Results

The PASCAL visual object classes challenge 2006 (VOC2006) results

Mislabeled

We found that the common ImageNet statistics normalization [14, 7, 26] does not improve results, and instead used a simpler normalization -scaling all the RGB channels to be between 0 and 1

On All Common Metrics In Table 10 we compare ASL results, to known stateof-the-art methods

However, using Higher input resolutions with larger architectures is highly beneficial for the mAP score. Notice that the input resolution is not always mentioned in articles

Soft-threshold - negative samples, with p > m , that should be attenuated when their probability is low

Field of Study

Journal Information

Name

Page

Venue Information

Name

Type

URL

Alternate Names