How to use the scripts modelsummary.py to compute the HRNet 's imformation? about hrnet-image-classification

[monk42]

Thanks !

[wondervictor]

Summary of parameters and GFLOPs will show before training, you can see it in train.py

HRNet-Image-Classification/tools/train.py

Line 90 in d698921

logger.info(get_model_summary(model, dump_input))

Or you can initialize your model and create an input like this:

# first you need define your model
x = torch.randn((1, 3, 224, 224))
print(get_model_summary(model, x))

[monk42]

Thanks ,Why computs for Convolution and Linear Layers only ? what about relu layer and bn layer?

[wondervictor]

When inference, BN can be combined with convolution. And GFLOPs of nonlinearity will not be involved in. For more details, you can refer to these papers: ShuffleNetV2, Pruning Convolutional Neural Networks for Resource Efficient Inference, etc.

[monk42]

Thanks！