Even today’s best #deeplearning vision systems can fail when pose/lighting/background vary. Existing benchmarks show challenging examples but don’t explain why. ImageNetX includes annotations for the entire ImageNet1k validation set + a random subset of 12k training images.

→ View original post on X — @aiatmeta,