Multiple GPU's and optimzer/lr_scheduler ordering

kensakurada / sscdnet

Semantic Scene Change Detection Network (CSCDNet + SSCDNet)

MIT License

106 stars 23 forks source link

Multiple GPU's and optimzer/lr_scheduler ordering #5

Open bigrobinson opened 4 years ago

bigrobinson commented 4 years ago

Enable training with multiple GPU's as well as saving model generically when using dataparallel. Also call lr_scheduler after optimizer to bring in line with Pytorch 1.4 requirements.