Ieee Transactions on Pattern Analysis and Machine Intelligence 1 80 Million Tiny Images: a Large Dataset for Non-parametric Object and Scene Recognition

By Antonio Torralba, Rob Fergus, William T. Freeman

Subfields: Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

ImageNet Large Scale Visual Recognition Challenge (2014, methodology): “The TinyImages dataset (Torralba et al. 2008) contains 80 million 32× 32 low resolution images collected from the internet using synsets in WordNet (Miller 1995) as queries.”
Learning Transferable Visual Models From Natural Language Supervision (2021, background): “It is clear, however, what distribution CIFAR-10 is drawn from - TinyImages (Torralba et al., 2008).”
Rethinking Few-Shot Image Classification: a Good Embedding Is All You Need? (2020, methodology): “benchmarks: miniImageNet [55], tieredImageNet [42], CIFAR-FS [3], and FC100 [34]. The ﬁrst two are derivatives of ImageNet [43], while the last two are reorganized from the standard CIFAR-100 dataset [23,52]. Additional results on Meta-Dataset [54] is presented in x5. 4.1. Setup Architecture. Follow”
Do ImageNet Classifiers Generalize to ImageNet? (2019, methodology): “The dataset contains 80 million RGB color images with resolution 32 × 32 pixels and was released in 2007 [55].”
A Study and Comparison of Human and Deep Learning Recognition Performance under Visual Distortions (2017, background): “[10] show that human subjects can accurately recognize low resolution images.”
Spectral Normalization for Generative Adversarial Networks (2018, methodology): “Table 3: Standard CNN models for CIFAR-10 and STL-10 used in our experiments on image Generation.”
SUN database: Large-scale scene recognition from abbey to zoo (2010, methodology): “For each scene category, images were retrieved using WordNet terminology from various search engines on the web [28].”
Towards fairer datasets: filtering and balancing the distribution of the people subtree in the ImageNet hierarchy (2019, background): “[73], image search engines were only about 10% accurate, and thus a manual cleanup of the candidate images is needed.”
Temporal Ensembling for Semi-Supervised Learning (2016, methodology): “We ran two additional tests using unlabeled extra data from Tiny Images dataset (Torralba et al., 2008): one with randomly selected 500k extra images, most not corresponding to any of the CIFAR-100 categories, and another with a restricted set of 237k images from the categories that correspond to…”
HMDB: A large video database for human motion recognition (2011, background): “Current action recognition databases contain on the order of ten different action categories collected under fairly controlled conditions.”
Data and its (dis)contents: A survey of dataset development and use in machine learning research (2020, neutral): “Similarly, Prabhu and Birhane’s [2020] examination prompted the complete removal of the TinyImages dataset [Torralba et al., 2008].”
Do CIFAR-10 Classifiers Generalize to CIFAR-10? (2018, background): “Impor-tantly, CIFAR-10 draws from the larger Tiny Images repository that has signiﬁcantly more ﬁne-grained labels [19].”
Do Deep Nets Really Need to be Deep? (2013, background): “of a set of natural images from 10 different object classes: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck. The dataset is a labeled subset of the 80 million tiny images dataset[18] and is divided into 50,000 train and 10,000 test images. Each image is 32x32 pixels in 3 color ”
Image-Based Recommendations on Styles and Substitutes (2015, background): “What distinguishes the approach we propose here, however, is the fact that it succeeds despite the indirect-ness of the connection between the dataset and the quantity we hope to model.”