MLPerf ResNet-50 pre-trained model

vision
#1

Hi there,

I wonder, have someone migrated MLPerf ResNet-50 checkpoint to PyTorch?
It achieves top-1 accuracy of 76.456%, and it will help me a lot having that checkpoint as a baseline.

https://github.com/mlperf/inference/tree/master/v0.5/classification_and_detection

Thanks!

#2

I don’t know, how these mlperf models are stored or used, but NVIDIA provides Resnet50v1.5 with a similar or higher Top1 accuracy here.

#3

Thanks! I will definitely check it out.

While we’re at it, does PyTorch has something for MobileNet-V1?
Again, MLPerf offers one with top-1 accuracy of 71.676%, but I can’t find any pre-trained model for PyTorch with this accuracy. Any suggestions?

And thanks again!

#4

If someone will come across this post, I managed to migrate the weights from MLPerf to PyTorch models. They can be found here: https://github.com/gilshm/mlperf-pytorch