Webly Supervised Learning of Convolutional Networks

Abstract

We present an approach to utilize large amounts of web data for learning CNNs. Specifically inspired by curriculum learning, we present a two-step approach for CNN training. First, we use easy images to train an initial visual representation. We then use this initial CNN and adapt it to harder, more realistic images by leveraging the structure of data and categories. We demonstrate that our two-stage CNN outperforms a fine-tuned CNN trained on ImageNet on Pascal VOC 2012. We also demonstrate the strength of webly supervised learning by localizing objects in web images and training a R-CNN style detector. It achieves the best performance on VOC 2007 where no VOC training data is used. Finally, we show our approach is quite robust to noise and performs comparably even when we use image search results from March 2013 (pre-CNN image search era).

References in corpus (9)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Caffe: Convolutional Architecture for Fast Feature Embedding
• Going Deeper with Convolutions
• Training Deep Neural Networks on Noisy Labels with Bootstrapping
• Training Convolutional Networks with Noisy Labels
• Weakly- and Semi-Supervised Learning of a DCNN for Semantic Image Segmentation
• Fully Convolutional Multi-Class Multiple Instance Learning
• Simultaneous Detection and Segmentation
• Analyzing the Performance of Multilayer Neural Networks for Object Recognition

Cited by in corpus (50)

• Generalized Out-of-Distribution Detection: A Survey
• WebVision Database: Visual Learning and Understanding from Web Data
• Revisiting Unreasonable Effectiveness of Data in Deep Learning Era
• Domain Adaptation for Visual Applications: A Comprehensive Survey
• Looking Beyond Appearances: Synthetic Training Data for Deep CNNs in Re-identification
• Learning with Noise: Enhance Distantly Supervised Relation Extraction with Dynamic Transition Matrix
• Learning from Noisy Labels with Distillation
• Learning From Noisy Large-Scale Datasets With Minimal Supervision
• Love Thy Neighbors: Image Annotation by Exploiting Image Metadata
• CleanNet: Transfer Learning for Scalable Image Classifier Training with Label Noise
• Iterative Learning with Open-set Noisy Labels
• Weakly Supervised Semantic Segmentation using Web-Crawled Videos
• The Unreasonable Effectiveness of Noisy Data for Fine-Grained Recognition
• WebVision Challenge: Visual Learning and Understanding With Web Data
• Deep Learning Logo Detection with Data Expansion by Synthesising Context
• Scalable Deep Learning Logo Detection
• Unsupervised Category Discovery via Looped Deep Pseudo-Task Optimization Using a Large Scale Radiology Image Database
• A Semi-Supervised Two-Stage Approach to Learning from Noisy Labels
• Learning Visual Features from Large Weakly Supervised Data
• Contrastive Learning Improves Model Robustness Under Label Noise
• Learning to Learn from Noisy Web Videos
• Active Mini-Batch Sampling using Repulsive Point Processes
• Unsupervised Joint Mining of Deep Features and Image Labels for Large-scale Radiology Image Categorization and Scene Recognition
• Mining Object Parts from CNNs via Active Question-Answering
• Hard Mixtures of Experts for Large Scale Weakly Supervised Vision
• Modeling Multimodal Clues in a Hybrid Deep Learning Framework for Video Classification
• Learning without Prejudice: Avoiding Bias in Webly-Supervised Action Recognition
• Commonly Uncommon: Semantic Sparsity in Situation Recognition
• Attend in groups: a weakly-supervised deep learning framework for learning from web data
• Exploiting Web Images for Weakly Supervised Object Detection
• Multimodal Co-Training for Selecting Good Examples from Webly Labeled Video
• Improving Human Activity Recognition Through Ranking and Re-ranking
• Zero-Annotation Object Detection with Web Knowledge Transfer
• Deep Learning from Noisy Image Labels with Quality Embedding
• Learning from Web Data: the Benefit of Unsupervised Object Localization
• Exploiting Multi-modal Curriculum in Noisy Web Data for Large-scale Concept Learning
• Learning from Noisy Web Data with Category-level Supervision
• DeeSIL: Deep-Shallow Incremental Learning
• Complex Event Recognition from Images with Few Training Examples
• Graph Distillation for Action Detection with Privileged Modalities
• Contemplating Visual Emotions: Understanding and Overcoming Dataset Bias
• Privacy-preserving Stochastic Gradual Learning
• Attention Transfer from Web Images for Video Recognition
• Combining Weakly and Webly Supervised Learning for Classifying Food Images
• Instance-Level Salient Object Segmentation
• Weakly Supervised Domain-Specific Color Naming Based on Attention
• Mining Interpretable AOG Representations from Convolutional Networks via Active Question Answering
• Unsupervised Multi-label Dataset Generation from Web Data
• Fine-grained Classification using Heterogeneous Web Data and Auxiliary Categories
• Training Convolutional Networks with Web Images