Measuring Robustness to Natural Distribution Shifts in Image Classification
This work highlights a critical gap in AI robustness for real-world image classification applications, showing that current methods are largely ineffective against natural distribution shifts.
The study evaluated 204 ImageNet models across 213 test conditions to assess robustness to natural distribution shifts, finding little transfer from synthetic to natural shifts and that most techniques fail to improve robustness, with only training on larger datasets offering limited gains.
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/ .