hcw-00/GroupNorm-reproduce

An official collection of code in different frameworks that reproduces experiments in "Group Normalization"

A collection of verified code in different frameworks, that reproduces several experiment results in the paper:

• Group Normalization - Best Paper Honorable Mention at ECCV 2018.

Reproduced Experiments from the Paper:

• Object Detection with Mask R-CNN, in Caffe2, with models. Official models for the paper.

• ImageNet Classification with ResNet-50, in TensorFlow, with models. Exactly reproduce.

• ImageNet Classification with ResNet-50, in PyTorch, with models. 0.1% different result due to fewer augmentations.

• ImageNet Classification with VGG-16, in TensorFlow. Exactly reproduce.

• Object Detection with Mask R-CNN, in TensorFlow, with models. Slighly better results due to different implementation of Mask R-CNN framework.

• Object Detection with Mask R-CNN, in PyTorch, with models. Slighly better results due to different implementation of Mask R-CNN framework.

Not in the Paper

• Cifar10 Classification with ResNet, in TensorFlow.

Implementations of the GroupNorm operation:

• PyTorch: Implemented in C++, and available as a layer via torch.nn.GroupNorm.
• TensorFlow: Python implementation.
• Caffe2: Implemented in C++, available in Python via brew.spatial_gn.