Question about SyncBN of Mutil-GPU

HRNet / HRNet-Object-Detection

Object detection with multi-level representations generated from deep high-resolution representation learning (HRNetV2h). This is an official implementation for our TPAMI paper "Deep High-Resolution Representation Learning for Visual Recognition". https://arxiv.org/abs/1908.07919

Apache License 2.0

643 stars 98 forks source link

Question about SyncBN of Mutil-GPU #10

Closed xunhen closed 5 years ago

xunhen commented 5 years ago

Hello! Did you use syncBN during training this HRNet? I find that the mmdetection had not achieved the function of syncBN in the version you are using.

wondervictor commented 5 years ago

Well, our released code does not support SyncBN. All models are trained with fixed batch normalization. If you want to train HRNets with SyncBN, using SyncBN in PyTorch 1.1 might be a great choice. (Just changing all BatchNorm into SyncBatchNorm)

xunhen commented 5 years ago

Thank you very much!