Learning Deep Features for Discriminative Localization

Abstract

In this work, we revisit the global average pooling layer proposed in [13], and shed light on how it explicitly enables the convolutional neural network to have remarkable localization ability despite being trained on image-level labels. While this technique was previously proposed as a means for regularizing training, we find that it actually builds a generic localizable deep representation that can be applied to a variety of tasks. Despite the apparent simplicity of global average pooling, we are able to achieve 37.1% top-5 error for object localization on ILSVRC 2014, which is remarkably close to the 34.2% top-5 error achieved by a fully supervised CNN approach. We demonstrate that our network is able to localize the discriminative image regions on a variety of tasks despite not being trained for them

References in corpus (2)

• Going Deeper with Convolutions
• Simple Baseline for Visual Question Answering

Cited by in corpus (75)

• Simple Baseline for Visual Question Answering
• A feature agnostic approach for glaucoma detection in OCT volumes
• Deformable ConvNets v2: More Deformable, Better Results
• ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression
• What Do We Understand About Convolutional Networks?
• Weakly Supervised Medical Diagnosis and Localization from Multiple Resolutions
• Towards Interpretable Deep Neural Networks by Leveraging Adversarial Examples
• Parameter-Free Spatial Attention Network for Person Re-Identification
• Adversarial Complementary Learning for Weakly Supervised Object Localization
• Two-Stream Neural Networks for Tampered Face Detection
• Morphological classification of galaxies with deep learning: comparing 3-way and 4-way CNNs
• Weakly Supervised Action Localization by Sparse Temporal Pooling Network
• A Taxonomy and Library for Visualizing Learned Features in Convolutional Neural Networks
• Knowing When to Look: Adaptive Attention via A Visual Sentinel for Image Captioning
• SS-CAM: Smoothed Score-CAM for Sharper Visual Feature Localization
• Large-Scale Image Retrieval with Attentive Deep Local Features
• Time Series Classification from Scratch with Deep Neural Networks: A Strong Baseline
• VideoMix: Rethinking Data Augmentation for Video Classification
• Prospects for Theranostics in Neurosurgical Imaging: Empowering Confocal Laser Endomicroscopy Diagnostics via Deep Learning
• ContextLocNet: Context-Aware Deep Network Models for Weakly Supervised Localization
• Two-Phase Learning for Weakly Supervised Object Localization
• Classifying Exoplanet Candidates with Convolutional Neural Networks: Application to the Next Generation Transit Survey
• Chester: A Web Delivered Locally Computed Chest X-Ray Disease Prediction System
• Weakly-Supervised Semantic Segmentation by Iteratively Mining Common Object Features
• Oriented Response Networks
• IS-CAM: Integrated Score-CAM for axiomatic-based explanations
• It Takes Two to Tango: Towards Theory of AI's Mind
• Robust Tumor Localization with Pyramid Grad-CAM
• Action Machine: Rethinking Action Recognition in Trimmed Videos
• Attention-based Context Aggregation Network for Monocular Depth Estimation
• Meta-DETR: Image-Level Few-Shot Object Detection with Inter-Class Correlation Exploitation
• Teaching Categories to Human Learners with Visual Explanations
• Interpretable Convolutional Neural Networks
• CheXpedition: Investigating Generalization Challenges for Translation of Chest X-Ray Algorithms to the Clinical Setting
• Weakly-supervised Visual Grounding of Phrases with Linguistic Structures
• Cardiac Segmentation on CT Images through Shape-Aware Contour Attentions
• Improving Interpretability of Deep Neural Networks with Semantic Information
• Interactively Transferring CNN Patterns for Part Localization
• xCos: An Explainable Cosine Metric for Face Verification Task
• Visual Relationship Detection using Scene Graphs: A Survey
• Learning a Discriminative Filter Bank within a CNN for Fine-grained Recognition
• ProNet: Learning to Propose Object-specific Boxes for Cascaded Neural Networks
• Mining Object Parts from CNNs via Active Question-Answering
• Interpretation of Deep Temporal Representations by Selective Visualization of Internally Activated Nodes
• Improving Object Detection with Inverted Attention
• Explaining Predictions by Approximating the Local Decision Boundary
• Top-down Visual Saliency Guided by Captions
• From Shallow to Deep Interactions Between Knowledge Representation, Reasoning and Machine Learning (Kay R. Amel group)
• Visual Concept Recognition and Localization via Iterative Introspection
• Understanding Convolutional Networks with APPLE : Automatic Patch Pattern Labeling for Explanation
• Segmentation of Surgical Instruments for Minimally-Invasive Robot-Assisted Procedures Using Generative Deep Neural Networks
• Learning to discover and localize visual objects with open vocabulary
• Recurrent U-net: Deep learning to predict daily summertime ozone in the United States
• What and Where: A Context-based Recommendation System for Object Insertion
• Cube Padding for Weakly-Supervised Saliency Prediction in 360° Videos
• Evaluating the performance of the LIME and Grad-CAM explanation methods on a LEGO multi-label image classification task
• Efficient Image Evidence Analysis of CNN Classification Results
• Viraliency: Pooling Local Virality
• Tournament Based Ranking CNN for the Cataract grading
• When Differential Privacy Meets Interpretability: A Case Study
• Thoracic Disease Identification and Localization using Distance Learning and Region Verification
• Recurrent and Spiking Modeling of Sparse Surgical Kinematics
• Classification of Breast Cancer Lesions in Ultrasound Images by using Attention Layer and loss Ensembles in Deep Convolutional Neural Networks
• Single Sample Feature Importance: An Interpretable Algorithm for Low-Level Feature Analysis
• Dense xUnit Networks
• Fast and interpretable classification of small X-ray diffraction datasets using data augmentation and deep neural networks
• Weakly Supervised Foreground Learning for Weakly Supervised Localization and Detection
• DeepWheat: Estimating Phenotypic Traits from Crop Images with Deep Learning
• Beyond Attributes: Adversarial Erasing Embedding Network for Zero-shot Learning
• Train, Diagnose and Fix: Interpretable Approach for Fine-grained Action Recognition
• Egok360: A 360 Egocentric Kinetic Human Activity Video Dataset
• A Gated Peripheral-Foveal Convolutional Neural Network for Unified Image Aesthetic Prediction
• A Comparative Study on Effects of Original and Pseudo Labels for Weakly Supervised Learning for Car Localization Problem
• Self-Improving Semantic Perception for Indoor Localisation
• Scientific Calculator for Designing Trojan Detectors in Neural Networks