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

I.I.D.2 hypothesis between training and testing data is the basis of numerous image classification methods. Such property can hardly be guaranteed in practice where Non-IIDness common, causing instable performances these models. In literature, however, Non-I.I.D.3 problem largely understudied. A key reason lacking a well-designed dataset to support related research. this paper, we construct release Non-I.I.D. called NICO4, which uses contexts create consciously. Compared other datasets, extended analyses prove NICO various situations with sufficient flexibility. Meanwhile, propose baseline model ConvNet structure for General classification, distribution unknown but different from data. The experimental results demonstrate that well scratch, batch balancing module help ConvNets perform better settings.