WebVision Database: Visual Learning and Understanding from Web Data

By Wen Li, Limin Wang, Eirikur Agustsson, Luc Van Gool. 2017

515 citations recorded by the source corpus.

In this paper, we present a study on learning visual recognition models from large scale noisy web data. We build a new database called WebVision, which contains more than $2.4$ million web images crawled from the Internet by using queries generated from the 1,000 semantic concepts of the benchmark ILSVRC 2012 dataset. Meta information along with those web images (e.g., title, description, tags, etc.) are also crawled. A validation set and test set containing human annotated images are also provided to facilitate algorithmic development. Based on our new database, we obtain a few interesting observations: 1) the noisy web images are sufficient for training a good deep CNN model for visual recognition; 2) the model learnt from our WebVision database exhibits comparable or even better generalization ability than the one trained from the ILSVRC 2012 dataset when being transferred to new datasets and tasks; 3) a domain adaptation issue (a.k.a., dataset bias) is observed, which means the dataset can be used as the largest benchmark dataset for visual domain adaptation. Our new WebVision database and relevant studies in this work would benefit the advance of learning state-of-the-art visual models with minimum supervision based on web data.

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

Online Continual Learning on a Contaminated Data Stream with Blurry Task Boundaries (2022, methodology): “Comparing PuriDivER with other methods, in particular, accuracy increases up to 16% on CIFARs, 11% and 3% on WebVision and Food-101N, respectively.”
Continual Learning on Noisy Data Streams via Self-Purified Replay (2021, methodology): “We evaluate our approach on three synthetic noise benchmarks of MNIST [41], CIFAR-10 [40], CIFAR100 [40] and one real noise dataset of WebVision [49].”
Tencent ML-Images: A Large-Scale Multi-Label Image Database for Visual Representation Learning (2019, methodology): “Widely used single-label image databases include CIFAR-10 [8], Caltech-256 [9], MNIST [10], ImageNet [2], WebVision [11], SUN [12] and Places [13], etc.”
MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels (2017, methodology): “To verify MentorNet on real-world noisy labels, we conduct experiments on the large WebVision benchmark (Li et al., 2017a).”
Efficient Self-supervised Vision Transformers for Representation Learning (2021, background): “To push the frontier in leveraging large amounts of unlabeled data, we explore the effects of pre-training from larger, less-curated image datasets: WebVision-v1 [36], OpenImages-v4 [30] and ImageNet-22K [15], described in Appendix.”
MoPro: Webly Supervised Learning with Momentum Prototypes (2020, methodology): “Approximately 20% of the labels in the JMT-300M dataset (Sun et al., 2017) are noisy, whereas 34% of images in the WebVision dataset (Li et al., 2017) are considered outliers.”
Transferable Clean-Label Poisoning Attacks on Deep Neural Nets (2019, methodology): “Both the target fish image and the five hook images used for crafting poisons come from the WebVision (Li et al., 2017) dataset, which has the same taxonomy as the ImageNet dataset.”
Normalized Loss Functions for Deep Learning with Noisy Labels (2020, methodology): “Here, we test the effectiveness of our APL loss functions on large-scale real-world noisy dataset WebVision (Li et al., 2017a).”
DivideMix: Learning with Noisy Labels as Semi-supervised Learning (2020, methodology): “WebVision contains 2.4 million images crawled from the web using the 1,000 concepts in ImageNet ILSVRC12.”
Concept Generalization in Visual Representation Learning (2020, unknown): “As it is not clear if YFCC-100M, IG-1B, WebImageText or WebVision-V1 contain images of the unseen concepts we selected in the levels, models in this category are not directly comparable .”
Robust Learning by Self-Transition for Handling Noisy Labels (2020, background): “0 [21], real-world noisy data of large-scale web images crawled from Flickr and Google Images search; and FOOD-101N [17], realworld noisy data of crawled food images annotated by their search keywords in the Food-101 Taxonomy.”
Synthetic vs Real: Deep Learning on Controlled Noise (2019, methodology): “This data subset contains only the images from Google text-to-image search and hence is similar to the noisy datasets used in previous studies (Li et al., 2017a; Krause et al., 2016; Chen & Gupta, 2015) (note that these existing datasets do not have controlled noise).”
A Survey on Recent Approaches for Natural Language Processing in Low-Resource Scenarios (2020, background): “The existing work on other tasks and the popularity in other fields like image classification (Xiao et al., 2015; Li et al., 2017; Lee et al., 2018; Mahajan et al., 2018; Li et al., 2020) suggests, however, that distant supervision could”
Early-Learning Regularization Prevents Memorization of Noisy Labels (2020, methodology): “We evaluate the proposed methodology on two standard benchmarks with simulated label noise, CIFAR-10 and CIFAR-100 [18], and two real-world datasets, Clothing1M [47] and WebVision [24].”
Beyond Synthetic Noise: Deep Learning on Controlled Noisy Labels (2019, methodology): “This can be achieved by querying a search index (Krause et al., 2016; Li et al., 2017a; Mahajan et al., 2018).”
CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images (2018, methodology): “Image samples of the WebVision dataset [19] from the categories of Carton , Dog , Taxi and Banana .”
Learn from Others and Be Yourself in Heterogeneous Federated Learning (2022, background): “, existing datasets [8, 51, 69] and web images [44].”