search

bwconrad / decoder-denoising

Pytorch reimplementation of Decoder Denoising Pretraining for Semantic Segmentation
MIT License
43 stars 7 forks source link

Question about implementation details #1

Closed SonwYang closed 1 year ago

SonwYang commented 1 year ago

Hello, bro! Nice work! Can you tell me about the number of epoch in Denoising pretraining for decoder only?

bwconrad commented 1 year ago

According to the original paper, they pretrain for 100 epochs on ImageNet-21k.

SonwYang commented 1 year ago

Thank you! I have another question:What kind of learning rate schedule used in training model?

------------------ 原始邮件 ------------------ 发件人: "Ben @.>; 发送时间: 2022年8月4日(星期四) 下午5:59 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [bwconrad/decoder-denoising] Question about implementation details (Issue #1)

According to the original paper, they pretrain for 100 epochs on ImageNet-21k.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

bwconrad commented 1 year ago

They train following a cosine annealing schedule which is what is used in this repo as well.