Towards Non-I.I.D. image classification: A dataset and baselines

By Yue He, Zheyan Shen, Peng Cui. 2019

211 citations recorded by the source corpus.

I.I.D. hypothesis between training and testing data is the basis of numerous image classification methods. Such property can hardly be guaranteed in practice where the Non-IIDness is common, causing instable performances of these models. In literature, however, the Non-I.I.D. image classification problem is largely understudied. A key reason is lacking of a well-designed dataset to support related research. In this paper, we construct and release a Non-I.I.D. image dataset called NICO, which uses contexts to create Non-IIDness consciously. Compared to other datasets, extended analyses prove NICO can support various Non-I.I.D. situations with sufficient flexibility. Meanwhile, we propose a baseline model with ConvNet structure for General Non-I.I.D. image classification, where distribution of testing data is unknown but different from training data. The experimental results demonstrate that NICO can well support the training of ConvNet model from scratch, and a batch balancing module can help ConvNets to perform better in Non-I.I.D. settings.

Subfields: Artificial Intelligence, Neurology, Radiology, Nuclear Medicine and Imaging

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

1. Tao Qin
2. Aaron C. Courville
3. Yoshua Bengio
4. Luc Van Gool
5. Mehrtash Harandi
6. Yi-Hsuan Tsai
7. Hai Helen Li
8. Christian Simon

In Neurology

1. Peng Cui
2. Xiang Yu
3. Mehrtash Harandi
4. Manmohan Chandraker
5. Amy Zhang
6. Yi-Hsuan Tsai
7. Hai Helen Li
8. Yiran Chen

In Radiology, Nuclear Medicine and Imaging

1. Yi-Hsuan Tsai
2. Mehrtash Harandi
3. Christian Simon
4. Peng Cui
5. Xiang Yu
6. Aaron C. Courville
7. Binghui Wang
8. Masoud Faraki

Citations in the corpus

• On Generalizing Beyond Domains in Cross-Domain Continual Learning (2022, methodology): “In addition, we evaluate on the PACS and the NICO-Animal datasets.”
• WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, neutral): “For example, BREEDS (Santurkar et al., 2020) tests generalization to unseen ImageNet subclasses by holding out subclasses as specified by several controllable parameters; similarly, NICO (He et al., 2020) considers subclasses that are defined by their context, such as dogs at home versus dogs on the”
• Generalizing to Unseen Domains: A Survey on Domain Generalization (2021, methodology): “[136] reweighted input samples in a way to make the weighted correlation reflect causal effect.”
• Out-of-Distribution Generalization via Risk Extrapolation (REx) (2020, background): “Several recent works also use prior knowledge to design augmentation strategies for invariance to superficial features that may be spuriously correlated with labels in object recognition tasks (He et al., 2019; Wang et al., 2019; Gowal et al., 2019; Ilse et al., 2020).”
• LotteryFL: Personalized and Communication-Efficient Federated Learning with Lottery Ticket Hypothesis on Non-IID Datasets (2020, background): “[20] define a metric named Non-IID Index (NI) to quantify the distribution shift between training data and test data.”
• Deep Stable Learning for Out-Of-Distribution Generalization (2021, methodology): “We consider four datasets to carry through these four settings, namely PACS [31], VLCS [58], MNIST-M [15] and NICO [20].”