CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images

Abstract

We present a simple yet efficient approach capable of training deep neural networks on large-scale weakly-supervised web images, which are crawled raw from the Internet by using text queries, without any human annotation. We develop a principled learning strategy by leveraging curriculum learning, with the goal of handling a massive amount of noisy labels and data imbalance effectively. We design a new learning curriculum by measuring the complexity of data using its distribution density in a feature space, and rank the complexity in an unsupervised manner. This allows for an efficient implementation of curriculum learning on large-scale web images, resulting in a high-performance CNN model, where the negative impact of noisy labels is reduced substantially. Importantly, we show by experiments that those images with highly noisy labels can surprisingly improve the generalization capability of the model, by serving as a manner of regularization. Our approaches obtain state-of-the-art performance on four benchmarks: WebVision, ImageNet, Clothing-1M and Food-101. With an ensemble of multiple models, we achieved a top-5 error rate of 5.2% on the WebVision challenge for 1000-category classification. This result was the top performance by a wide margin, outperforming second place by a nearly 50% relative error rate. Code and models are available at: https://github.com/MalongTech/CurriculumNet .

arXiv commentAccepted to ECCV 2018. 16 pages, 5 figures, 5 tables

References in corpus (10)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• Focal Loss for Dense Object Detection
• DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs
• Deep Learning is Robust to Massive Label Noise
• WebVision Database: Visual Learning and Understanding from Web Data
• FoodNet: Recognizing Foods Using Ensemble of Deep Networks
• Locally-Supervised Deep Hybrid Model for Scene Recognition
• Learning From Noisy Large-Scale Datasets With Minimal Supervision
• WebVision Challenge: Visual Learning and Understanding With Web Data

Cited by in corpus (11)

• IMAE for Noise-Robust Learning: Mean Absolute Error Does Not Treat Examples Equally and Gradient Magnitude's Variance Matters
• Dynamic Curriculum Learning for Imbalanced Data Classification
• Density-Based Dynamic Curriculum Learning for Intent Detection
• Webly Supervised Image Classification with Metadata: Automatic Noisy Label Correction via Visual-Semantic Graph
• Weak-shot Fine-grained Classification via Similarity Transfer
• Grounding Object Detections With Transcriptions
• Weakly Supervised Learning with Side Information for Noisy Labeled Images
• Deep Learning Classification With Noisy Labels
• Safeguarded Dynamic Label Regression for Generalized Noisy Supervision
• Product Image Recognition with Guidance Learning and Noisy Supervision
• Unsupervised Multi-label Dataset Generation from Web Data