DeCAF: A Deep Convolutional Activation Feature for Generic Visual\n Recognition

2013/10/05 by Jeff Donahue, Donahue, Jeff, Yangqing Jia +11 · 100 citations
Computer Science · #Advanced Image and Video Retrieval Techniques #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Multimodal Machine Learning Applications

paper · pdf · doi:10.48550/arxiv.1310.1531

openalex publication_date 2013/10/05 · openalex created_date 2025/10/24 · openalex updated_date 2026/07/28

Abstract

We evaluate whether features extracted from the activation of a deep\nconvolutional network trained in a fully supervised fashion on a large, fixed\nset of object recognition tasks can be re-purposed to novel generic tasks. Our\ngeneric tasks may differ significantly from the originally trained tasks and\nthere may be insufficient labeled or unlabeled data to conventionally train or\nadapt a deep architecture to the new tasks. We investigate and visualize the\nsemantic clustering of deep convolutional features with respect to a variety of\nsuch tasks, including scene recognition, domain adaptation, and fine-grained\nrecognition challenges. We compare the efficacy of relying on various network\nlevels to define a fixed feature, and report novel results that significantly\noutperform the state-of-the-art on several important vision challenges. We are\nreleasing DeCAF, an open-source implementation of these deep convolutional\nactivation features, along with all associated network parameters to enable\nvision researchers to be able to conduct experimentation with deep\nrepresentations across a range of visual concept learning paradigms.\n

Citations

• Gradient-based learning applied to document recognition
• Backpropagation Applied to Handwritten Zip Code Recognition
• ImageNet Large Scale Visual Recognition Challenge

Cited by

• Greedy dynamical meta-learning
• Data-Centric Visual Development for Self-Driving Labs
• Bridging Lifelong and Multi-Task Representation Learning via Algorithm and Complexity Measure
• Learning an Efficient Optimizer via Hybrid-Policy Sub-Trajectory Balance
• Feature Stylization and Domain-aware Contrastive Learning for Domain Generalization
• Point Prompting: Counterfactual Tracking with Video Diffusion Models
• Mutual Learning for Hashing: Unlocking Strong Hash Functions from Weak Supervision
• Tent: Fully Test-time Adaptation by Entropy Minimization
• Accelerating Deep Neural Networks with Spatial Bottleneck Modules
• Unsupervised Domain Adaptation by Backpropagation
• Shared Neural Space: Unified Precomputed Feature Encoding for Multi-Task and Cross Domain Vision
• PAC-Bayes Analysis of Sentence Representation
• Learning to see across Domains and Modalities
• Leveraging Visual Question Answering for Image-Caption Ranking
• Convolutional Channel Features
• Weakly-Supervised Spatial Context Networks
• Online and Offline Handwritten Chinese Character Recognition: A Comprehensive Study and New Benchmark
• Context-Aware Multipath Networks
• Joint calibration of Ensemble of Exemplar SVMs
• From handcrafted to deep local features
• Make Your Bone Great Again : A study on Osteoporosis Classification
• Hierarchical Retrieval: The Geometry and a Pretrain-Finetune Recipe
• Domain Generalization via Multidomain Discriminant Analysis
• Online Learning to Sample
• Embedding Label Structures for Fine-Grained Feature Representation
• RCoNet: Deformable Mutual Information Maximization and High-order Uncertainty-aware Learning for Robust COVID-19 Detection
• One-Shot Imitation Learning
• Unsupervised Deep Domain Adaptation for Pedestrian Detection
• Is Meta-Learning Out? Rethinking Unsupervised Few-Shot Classification with Limited Entropy
• Temporal Attentive Alignment for Large-Scale Video Domain Adaptation
• Learning Transferable Features with Deep Adaptation Networks
• Learning Deep Representations of Fine-grained Visual Descriptions
• Bootstrap your own latent: A new approach to self-supervised Learning
• SpotTune: Transfer Learning through Adaptive Fine-tuning
• Recognizing American Sign Language Manual Signs from RGB-D Videos
• DISA at ImageCLEF 2014 Revised: Search-based Image Annotation with DeCAF Features
• Unlabeled Data Deployment for Classification of Diabetic Retinopathy Images Using Knowledge Transfer
• Latent Hinge-Minimax Risk Minimization for Inference from a Small Number of Training Samples
• Deep Learning for Generic Object Detection: A Survey
• CAGFuzz: Coverage-Guided Adversarial Generative Fuzzing Testing of Deep Learning Systems
• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• An Extended Framework for Marginalized Domain Adaptation
• Strength from Weakness: Fast Learning Using Weak Supervision
• Analysing domain shift factors between videos and images for object\n detection
• Transferring GANs: generating images from limited data
• Deep Learning Face Attributes in the Wild
• Universality of Gradient Descent Neural Network Training
• Label Efficient Learning of Transferable Representations across Domains and Tasks
• CO-Optimal Transport
• LogME: Practical Assessment of Pre-trained Models for Transfer Learning
• Cross-convolutional-layer Pooling for Image Recognition
• Analyzing the Performance of Multilayer Neural Networks for Object\n Recognition
• Fine-grained pose prediction, normalization, and recognition
• Structured Sparse Convolutional Autoencoder
• Learning Optimal Linear Regularizers
• VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback
• MemeSequencer: Sparse Matching for Embedding Image Macros
• Regularization via Mass Transportation
• CoMET: A Contrastive-Masked Brain Foundation Model for Universal EEG Representation
• Brain-Inspired Deep Networks for Image Aesthetics Assessment
• Compositional Model based Fisher Vector Coding for Image Classification
• EndoNet: A Deep Architecture for Recognition Tasks on Laparoscopic Videos
• Contrastively Smoothed Class Alignment for Unsupervised Domain Adaptation
• Material Classification using Neural Networks
• Deep Transfer Learning with Joint Adaptation Networks
• STAGNet: A Spatio-Temporal Graph and LSTM Framework for Accident Anticipation
• A Unified Semantic Embedding: Relating Taxonomies and Attributes
• Discriminative Noise Robust Sparse Orthogonal Label Regression-based Domain Adaptation
• Agent-Centric Representations for Multi-Agent Reinforcement Learning
• Open Set Domain Adaptation for Image and Action Recognition
• Ranking and Tuning Pre-trained Models: A New Paradigm for Exploiting Model Hubs
• Domain Adaptation and Image Classification via Deep Conditional Adaptation Network
• NeMo: A Neuron-Level Modularizing-While-Training Approach for Decomposing DNN Models
• The Effectiveness of Instance Normalization: a Strong Baseline for Single Image Dehazing
• ConFoc: Content-Focus Protection Against Trojan Attacks on Neural Networks
• Continual Learning in Neural Networks
• Prototypical Cross-domain Self-supervised Learning for Few-shot Unsupervised Domain Adaptation
• Partial Optimal Transport with Applications on Positive-Unlabeled\n Learning
• Compression of Deep Neural Networks on the Fly
• An Analysis of Human-centered Geolocation
• Mind the Nuisance: Gaussian Process Classification using Privileged\n Noise
• DART: Domain-Adversarial Residual-Transfer Networks for Unsupervised Cross-Domain Image Classification
• Defending Against Adversarial Attacks Using Random Forests
• Recovering 6D Object Pose: A Review and Multi-modal Analysis
• Robust Visual Knowledge Transfer via EDA
• Filtered Manifold Alignment
• Tailoring: encoding inductive biases by optimizing unsupervised\n objectives at prediction time
• Deep Affordance-grounded Sensorimotor Object Recognition
• HyperSTAR: Task-Aware Hyperparameters for Deep Networks
• "Who is Driving around Me?" Unique Vehicle Instance Classification using Deep Neural Features
• GAN Memory with No Forgetting
• Fine-Grained Categorization via CNN-Based Automatic Extraction and Integration of Object-Level and Part-Level Features
• Video Summarization with Long Short-term Memory
• Learning Smooth Representation for Unsupervised Domain Adaptation
• Deep Reconstruction-Classification Networks for Unsupervised Domain Adaptation
• Rethinking Pre-training and Self-training
• Outlining the Design Space of Explainable Intelligent Systems for Medical Diagnosis
• Transferable Calibration with Lower Bias and Variance in Domain Adaptation
• On the Behavior of Convolutional Nets for Feature Extraction
• Double-level adversarial domain adaptation network for intelligent fault diagnosis

Related

• Deep Convolutional Neural Networks as Generic Feature Extractors
• Network Comparison Study of Deep Activation Feature Discriminability with Novel Objects
• CNN Features off-the-shelf: an Astounding Baseline for Recognition
• Factors of Transferability for a Generic ConvNet Representation
• A Comprehensive Survey of Convolutions in Deep Learning: Applications, Challenges, and Future Trends