search

fcdl94 / MiB

Official code for Modeling the Background for Incremental Learning in Semantic Segmentation https://arxiv.org/abs/2002.00718
MIT License
167 stars 44 forks source link

i want to use standard BN #48

Closed JIA-HONG-CHU closed 2 years ago

JIA-HONG-CHU commented 2 years ago

Hello, i want to use batch normalization instead of In-place ABN. What setting should I revise ?. I only revised norm_act to std in argparser.py. But i got error torch.nn.modules.module.ModuleAttributeError: 'BatchNorm2d' object has no attribute 'activation'. tks for help.

fcdl94 commented 2 years ago

Hi @JIA-HONG-CHU.

To use standard batch norm I suggest you use "ABN", which is the "activated Batch Norm" version, which encapsulates standard batch norm followed by leaky-relu. Please refer to the In-Place ABN library for the details.

JIA-HONG-CHU commented 2 years ago

ok, I modified the Pytorch BN2d to ABN and the code works fine. tks . I will close this issue. ;)