CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images

By Sheng Guo, Weilin Huang, Haozhi Zhang, Chenfan Zhuang, Dengke Dong, Matthew R. Scott, Dinglong Huang. 2018

383 citations recorded by the source corpus.

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 the 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 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 [18] 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.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

IL2M: Class Incremental Learning With Dual Memory (2019, methodology): “We will pursue work along the following lines: (1) test a constant complexity method such as IL2M for multitask IL to replicate real-life scenarios in which more diversified visual content is encountered, (2) enhance vanilla fine tuning by leveraging recent results which improve imbalanced learning ”
Unsupervised label noise modeling and loss correction (2019, background): “Similarly, (Guo et al., 2018) present a curriculum learning approach based on an unsupervised estimation on data complexity through its distribution in a feature space that beneﬁts from training with both clean and noisy samples.”
Webly Supervised Concept Expansion for General Purpose Vision Models (2022, methodology): “[24] Sheng Guo, Weilin Huang, Haozhi Zhang, Chenfan Zhuang, Dengke Dong, Matthew R.”
Deep learning with noisy labels: exploring techniques and remedies in medical image analysis (2019, methodology): “[133] proposed another method based on curriculum learning, named CurriculumNet, for training a model from”
Decoupling Category-wise Independence and Relevance with Self-attention for Multi-label Image Classification (2019, methodology): “With the development of weakly-supervised learning (WSL), a number of WSL-based methods [11, 12, 13] has obtained promising performance on multi-label image classification.”
Synthetic vs Real: Deep Learning on Controlled Noise (2019, background): “For example, Guo et al. (2018), Jiang et al. (2018) and Song et al. (2018) veriﬁed their model on WebVision. et al., 2017) and other regularization techniques (Azadi et al., 2016; Noh et al., 2017), label cleaning/correction (Reed et al., 2014; Goldberger & Ben-Reuven, 2017; Li et al., 2017b; Veit…”
Part-dependent Label Noise: Towards Instance-dependent Label Noise (2020, background): “Learning with noisy labels can be dated back to [4], which has recently drawn a lot of attention, especially from the deep learning community, e.g., [52, 79, 25, 13, 50, 57, 77, 37, 70, 75, 18, 41, 53, 22, 40, 56, 17, 16, 61, 60, 33, 32, 31, 21, 39, 46, 72, 66, 78].”
Suppressing Uncertainties for Large-Scale Facial Expression Recognition (2020, result): “We also compare our SCN to two state-of-the-art noise-tolerant meth-ods on RAF-DB, namely CurriculumNet [14] and Meta-Cleaner [46].”
Beyond Synthetic Noise: Deep Learning on Controlled Noisy Labels (2019, methodology): “Mixup (Zhang et al., 2018) minimizes the empirical vicinal risk calculated from: ˜ x and ˜ y are computed by the mixup function: where λ is drawn from the Beta distribution Beta ( α, α ) controlled by hyperparameter α .”
MoPro: Webly Supervised Learning with Momentum Prototypes (2020, background): “There exists a large body of literature on learning with label noise (Jiang et al., 2018; Han et al., 2018; Guo et al., 2018; Tanaka et al., 2018; Arazo et al., 2019; Li et al., 2020a).”
Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision (2023, background): “Weak-to-strong learning is a special type of weakly supervised learning—a setting in which models are trained using unreliable labels (Bach et al., 2017; Ratner et al., 2017; Guo et al., 2018).”
Self-paced Contrastive Learning with Hybrid Memory for Domain Adaptive Object Re-ID (2020, background): “The “easy-to-hard” training scheme is at the core of self-paced learning [19], which was originally found effective in supervised learning methods, especially with noisy labels [11, 16, 23].”
Grounding Object Detections With Transcriptions (2019, methodology): “For this reason, to overcome this difficulty, there has been numerous works on weakly-supervised models for object and event classification (Zhuang et al., 2017; Guo et al., 2018; Yu et al., 2014; Chang et al., 2015).”