A few cells are empty due to resource constraints. Testbed code and data is provided at https://modestyachts.github.io/imagenet-testbed/. While larger training subsets yield higher accuracies, they do not improve effective robustness,atleastforImageNetV2. Wethenevaluateallmodels onthe125classsubset and showthe results in Figure 7. Varying the number of classes again affects accuracies, but does not impacteffectiverobustness. The models can be directly compared with each other since the base model before intervention is the same.

We study how robust current ImageNet models are to distribution shifts arising from natural variations in datasets. Most research on robustness focuses on synthetic image perturbations (noise, simulated weather artifacts, adversarial examples, etc.), which leaves open how robustness on synthetic distribution shift relates to distribution shift arising in real data. Informed by an evaluation of 204 ImageNet models in 213 different test conditions, we find that there is often little to no transfer of robustness from current synthetic to natural distribution shift. Moreover, most current techniques provide no robustness to the natural distribution shifts in our testbed. The main exception is training on larger and more diverse datasets, which in multiple cases increases robustness, but is still far from closing the performance gaps. Our results indicate that distribution shifts arising in real data are currently an open research problem. We provide our testbed and data as a resource for future work at https://modestyachts.github.io/imagenet-testbed/ .