Learning Deep Features for Discriminative Localization

2015/12/14 by Bolei Zhou, Zhou, Bolei, Aditya Khosla +7 · 143 citations
Computer Science · #Advanced Neural Network Applications #Advanced Image and Video Retrieval Techniques #Domain Adaptation and Few-Shot Learning

paper · pdf · doi:10.48550/arxiv.1512.04150

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

Cited by

• Cascade Forgery Mining Network for Fingerprint Presentation Attack Detection
• Gradient-Based Latent Decomposition Reveals Mechanisms of Feature Degradation in Weakly Supervised Mammography
• Same Predictions, Different Reasons: The Effect of Quantization on Model Explanations
• A Diagnostic Gap Framework for Evaluating Reconstruction Fidelity in Weakly Supervised Mammography
• Steering Vision-Language Pre-trained Models for Incremental Face Presentation Attack Detection
• Interpretable Plant Leaf Disease Detection Using Attention-Enhanced CNN
• Beyond Occlusion: In Search for Near Real-Time Explainability of CNN-Based Prostate Cancer Classification
• CLARiTy: A Vision Transformer for Multi-Label Classification and Weakly-Supervised Localization of Chest X-ray Pathologies
• SynthSeg-Agents: Multi-Agent Synthetic Data Generation for Zero-Shot Weakly Supervised Semantic Segmentation
• TWLR: Text-Guided Weakly-Supervised Lesion Localization and Severity Regression for Explainable Diabetic Retinopathy Grading
• Back to the Baseline: Examining Baseline Effects on Explainability Metrics
• Hybrid Transformer-Mamba for Weakly Supervised Volumetric Medical Segmentation
• Visual Categorization Across Minds and Models: Cognitive Analysis of Human Labeling and Neuro-Symbolic Integration
• Refining Visual Artifacts in Diffusion Models via Explainable AI-based Flaw Activation Maps
• Liver Fibrosis Quantification and Analysis: The LiQA Dataset and Baseline Method
• CAuSE: Decoding Multimodal Classifiers using Faithful Natural Language Explanation
• Always Keep Your Promises: A Model-Agnostic Attribution Algorithm for Neural Networks
• CLUENet: Cluster Attention Makes Neural Networks Have Eyes
• QGShap: Quantum Acceleration for Faithful GNN Explanations
• SSR: Semantic and Spatial Rectification for CLIP-based Weakly Supervised Segmentation
• A Self-explainable Model of Long Time Series by Extracting Informative Structured Causal Patterns
• Attention Trajectories as a Diagnostic Axis for Deep Reinforcement Learning
• Functional Localization Enforced Deep Anomaly Detection Using Fundus Images
• Lightweight Transformer Framework for Weakly Supervised Semantic Segmentation
• Investigating self-supervised representations for audio-visual deepfake detection
• Enhancing Adversarial Transferability through Block Stretch and Shrink
• Semi-Supervised Multi-Task Learning for Interpretable Quality As- sessment of Fundus Images
• Counterfactual Explainable AI (XAI) Method for Deep Learning-Based Multivariate Time Series Classification
• Rethinking Saliency Maps: A Cognitive Human Aligned Taxonomy and Evaluation Framework for Explanations
• H-CNN-ViT: A Hierarchical Gated Attention Multi-Branch Model for Bladder Cancer Recurrence Prediction
• SAGE: Saliency-Guided Contrastive Embeddings
• Did Models Sufficient Learn? Attribution-Guided Training via Subset-Selected Counterfactual Augmentation
• A Closer Look at Knowledge Distillation in Spiking Neural Network Training
• From Linear Probing to Joint-Weighted Token Hierarchy: A Foundation Model Bridging Global and Cellular Representations in Biomarker Detection
• Saliency-Guided Domain Adaptation for Left-Hand Driving in Autonomous Steering
• Beyond Softmax: Dual-Branch Sigmoid Architecture for Accurate Class Activation Maps
• Exploring Object-Aware Attention Guided Frame Association for RGB-D SLAM
• Two-Phase Learning for Weakly Supervised Object Localization
• Cube Padding for Weakly-Supervised Saliency Prediction in 360° Videos
• Oriented Response Networks
• Egok360: A 360 Egocentric Kinetic Human Activity Video Dataset
• Learning to discover and localize visual objects with open vocabulary
• Beyond Attributes: Adversarial Erasing Embedding Network for Zero-shot Learning
• Region-CAM: Towards Accurate Object Regions in Class Activation Maps for Weakly Supervised Learning Tasks
• HiMAE: Hierarchical Masked Autoencoders Discover Resolution-Specific Structure in Wearable Time Series
• HistoLens: An Interactive XAI Toolkit for Verifying and Mitigating Flaws in Vision-Language Models for Histopathology
• What Do We Understand About Convolutional Networks?
• Caption-Driven Explainability: Probing CNNs for Bias via CLIP
• BioCAP: Exploiting Synthetic Captions Beyond Labels in Biological Foundation Models
• Explainable Face Presentation Attack Detection via Ensemble-CAM
• 3D Weakly Supervised Semantic Segmentation via Class-Aware and Geometry-Guided Pseudo-Label Refinement
• Split-Fuse-Transport: Annotation-Free Saliency via Dual Clustering and Optimal Transport Alignment
• Simple Baseline for Visual Question Answering
• Closed-Loop Transfer for Weakly-supervised Affordance Grounding
• WP-CrackNet: A Collaborative Adversarial Learning Framework for End-to-End Weakly-Supervised Road Crack Detection
• Controlling the image generation process with parametric activation functions
• DCMIL: A Progressive Representation Learning of Whole Slide Images for Cancer Prognosis Analysis
• XD-RCDepth: Lightweight Radar-Camera Depth Estimation with Explainability-Aligned and Distribution-Aware Distillation
• Machine Learning Frameworks for Large-Scale Radio Surveys: A Summary of Recent Studies
• ViConEx-Med: Visual Concept Explainability via Multi-Concept Token Transformer for Medical Image Analysis
• Training-Free In-Context Forensic Chain for Image Manipulation Detection and Localization
• RASALoRE: Region Aware Spatial Attention with Location-based Random Embeddings for Weakly Supervised Anomaly Detection in Brain MRI Scans
• Enhancing Concept Localization in CLIP-based Concept Bottleneck Models
• Cluster Paths: Navigating Interpretability in Neural Networks
• MedCLM: Learning to Localize and Reason via a CoT-Curriculum in Medical Vision-Language Models
• CoPA: Hierarchical Concept Prompting and Aggregating Network for Explainable Diagnosis
• Onto-Epistemological Analysis of AI Explanations
• Provenance Networks: End-to-End Exemplar-Based Explainability
• TextCAM: Explaining Class Activation Map with Text
• What and Where: A Context-based Recommendation System for Object Insertion
• Teaching Categories to Human Learners with Visual Explanations
• Fast and accurate explanations of distance-based classifiers by uncovering latent explanatory structures
• CE-FAM: Concept-Based Explanation via Fusion of Activation Maps
• EVO-LRP: Evolutionary Optimization of LRP for Interpretable Model Explanations
• MindCraft: How Concept Trees Take Shape In Deep Models
• Visual Relationship Detection using Scene Graphs: A Survey
• Explaining multimodal LLMs via intra-modal token interactions
• FreqDebias: Towards Generalizable Deepfake Detection via Consistency-Driven Frequency Debiasing
• Learning to Look: Cognitive Attention Alignment with Vision-Language Models
• MoTiC: Momentum Tightness and Contrast for Few-Shot Class-Incremental Learning
• Achieving Fair Skin Lesion Detection through Skin Tone Normalization and Channel Pruning
• Understanding plant phenotypes in crop breeding through explainable AI
• Generative Feature Imputing -- A Technique for Error-resilient Semantic Communication
• xCos: An Explainable Cosine Metric for Face Verification Task
• Train, Diagnose and Fix: Interpretable Approach for Fine-grained Action Recognition
• AI-CNet3D: An Anatomically-Informed Cross-Attention Network with Multi-Task Consistency Fine-tuning for 3D Glaucoma Classification
• Debugging Concept Bottleneck Models through Removal and Retraining
• Assessing the Alignment of Popular CNNs to the Brain for Valence Appraisal
• Explainable Graph Neural Networks: Understanding Brain Connectivity and Biomarkers in Dementia
• xAI-CV: An Overview of Explainable Artificial Intelligence in Computer Vision
• Rational Multi-Modal Transformers for TCR-pMHC Prediction
• Losing the Plot: How VLM responses degrade on imperfect charts
• Looking in the mirror: A faithful counterfactual explanation method for interpreting deep image classification models
• IPF-RDA: An Information-Preserving Framework for Robust Data Augmentation
• CAMBench-QR : A Structure-Aware Benchmark for Post-Hoc Explanations with QR Understanding
• Benchmarking Class Activation Map Methods for Explainable Brain Hemorrhage Classification on Hemorica Dataset
• Attention Lattice Adapter: Visual Explanation Generation for Visual Foundation Model
• Unleashing the Potential of Multimodal LLMs for Zero-Shot Spatio-Temporal Video Grounding
• GW-YOLO: Multi-transient segmentation in LIGO using computer vision
• Advancing Weakly-Supervised Change Detection in Satellite Images via Adversarial Class Prompting
• CheXpedition: Investigating Generalization Challenges for Translation of\n Chest X-Ray Algorithms to the Clinical Setting
• Localized Region Guidance for Class Activation Mapping in WSSS
• CUBE: Contrastive Understanding by Balanced Experiments
• Self-supervised Learning Of Visual Pose Estimation Without Pose Labels By Classifying LED States
• Dual-Thresholding Heatmaps to Cluster Proposals for Weakly Supervised Object Detection
• Object-level Correlation for Few-Shot Segmentation
• MAIA: An Inpainting-Based Approach for Music Adversarial Attacks
• Thoracic Disease Identification and Localization using Distance Learning and Region Verification
• MCANet: A Multi-Scale Class-Specific Attention Network for Multi-Label Post-Hurricane Damage Assessment using UAV Imagery
• Ecologically Valid Benchmarking and Adaptive Attention: Scalable Marine Bioacoustic Monitoring
• Beyond Output Faithfulness: Learning Attributions that Preserve Computational Pathways
• Invariant Features for Global Crop Type Classification
• DRetNet: A Novel Deep Learning Framework for Diabetic Retinopathy Diagnosis
• Pose as Clinical Prior: Learning Dual Representations for Scoliosis Screening
• The Demon is in Ambiguity: Revisiting Situation Recognition with Single Positive Multi-Label Learning
• Safer Skin Lesion Classification with Global Class Activation Probability Map Evaluation and SafeML
• A Novel Framework for Automated Explain Vision Model Using Vision-Language Models
• Using Machine Learning to Generate, Clarify, and Improve Economic Models
• Explainable Knowledge Distillation for Efficient Medical Image Classification
• AT-CXR: Uncertainty-Aware Agentic Triage for Chest X-rays
• No Label Left Behind: A Unified Surface Defect Detection Model for all Supervision Regimes
• SWiFT: Soft-Mask Weight Fine-tuning for Bias Mitigation
• Large-Scale Image Retrieval with Attentive Deep Local Features
• Breakable Machine: A K-12 Classroom Game for Transformative AI Literacy Through Spoofing and eXplainable AI (XAI)
• Genuine multipartite entanglement verification with convolutional neural networks
• Neural Cellular Automata for Weakly Supervised Segmentation of White Blood Cells
• PCA- and SVM-Grad-CAM for Convolutional Neural Networks: Closed-form Jacobian Expression
• Deep Learning for Crack Detection: A Review of Learning Paradigms, Generalizability, and Datasets
• Leveraging Failed Samples: A Few-Shot and Training-Free Framework for Generalized Deepfake Detection
• Selective Contrastive Learning for Weakly Supervised Affordance Grounding
• Attribution Explanations for Deep Neural Networks: A Theoretical Perspective
• MIMIC: Multimodal Inversion for Model Interpretation and Conceptualization
• SynSeg: Feature Synergy for Multi-Category Contrastive Learning in End-to-End Open-Vocabulary Semantic Segmentation
• WSS-CL: Weight Saliency Soft-Guided Contrastive Learning for Efficient Machine Unlearning Image Classification
• CORE-ReID: Comprehensive Optimization and Refinement through Ensemble fusion in Domain Adaptation for person re-identification
• Semi-Supervised Dual-Threshold Contrastive Learning for Ultrasound Image Classification and Segmentation
• DisFaceRep: Representation Disentanglement for Co-occurring Facial Components in Weakly Supervised Face Parsing
• Weakly Supervised Virus Capsid Detection with Image-Level Annotations in Electron Microscopy Images
• Training-Free Class Purification for Open-Vocabulary Semantic Segmentation
• Decouple before Align: Visual Disentanglement Enhances Prompt Tuning
• Explainable Image Classification with Reduced Overconfidence for Tissue Characterisation
• Mitigating Spurious Correlations in Weakly Supervised Semantic Segmentation via Cross-architecture Consistency Regularization
• ProNet: Learning to Propose Object-specific Boxes for Cascaded Neural Networks
• Towards trustworthy AI in materials mechanics through domain-guided attention
• Interpretation of Deep Temporal Representations by Selective Visualization of Internally Activated Nodes

Related

• Global Weighted Average Pooling Bridges Pixel-level Localization and Image-level Classification
• Self-taught Object Localization with Deep Networks
• Multi-scale discriminative Region Discovery for Weakly-Supervised Object Localization
• Latent Model Ensemble with Auto-localization
• OverFeat: Integrated Recognition, Localization and Detection using\n Convolutional Networks