Datasets:
Tasks:
Image Classification
Sub-tasks:
multi-class-image-classification
Languages:
English
Size:
100K - 1M
Tags:
adversarial-robustness
common-corruptions
robustness
benchmark
transfer-attacks
vision-language-models
License:
v2.1: common split from the original processed archives
#2
by legolasflagstaff - opened
Replaces the common split of all six datasets with the common-corruption images of the original processed archives (test_common/), which were used to evaluate the vision models in the paper. Clean, random, and adversarial splits are unchanged (byte-identical). v2.0 is tagged v2.0.
legolasflagstaff changed pull request status to merged