Thread by @paperswithcode

In a new paper from @wightmanr et al. a traditional ResNet-50 is re-trained using a modern training protocol. It achieves a very competitive 80.4% top-1 accuracy on ImageNet without using extra data or distillation.

[mini-thread]

paperswithcode.com/paper/resnet-s…

The paper catalogues the exact training settings to provide a robust baseline for future experiments:

It also records training costs and inference times on ImageNet classification between other architectures trained with the proposed ResNet-50 optimized training procedure:

It evaluates transfer learning performance using different procedures on a number of standard benchmarks. One of the proposed procedures (A1) leads to the best overall performance on the downstream tasks:

Code: github.com/rwightman/pyto…
Docs: rwightman.github.io/pytorch-image-…

For more comparisons visit the ImageNet benchmark: paperswithcode.com/sota/image-cla…

...and other image classification benchmarks:
paperswithcode.com/task/image-cla…

This is part of a new series showcasing new ML papers in the community. Stay tuned for more paper highlights.