Do ImageNet Classifiers Generalize to ImageNet?
This addresses a critical issue for machine learning researchers by revealing generalization gaps in widely used benchmarks, though it is incremental as it builds on existing dataset creation processes.
The paper tackled the problem of overfitting to re-used test sets in image classification by creating new test sets for CIFAR-10 and ImageNet, finding accuracy drops of 3-15% and 11-14% respectively, indicating models struggle to generalize to slightly harder images.
We build new test sets for the CIFAR-10 and ImageNet datasets. Both benchmarks have been the focus of intense research for almost a decade, raising the danger of overfitting to excessively re-used test sets. By closely following the original dataset creation processes, we test to what extent current classification models generalize to new data. We evaluate a broad range of models and find accuracy drops of 3% - 15% on CIFAR-10 and 11% - 14% on ImageNet. However, accuracy gains on the original test sets translate to larger gains on the new test sets. Our results suggest that the accuracy drops are not caused by adaptivity, but by the models' inability to generalize to slightly "harder" images than those found in the original test sets.