search

er-muyue / DeFRCN

MIT License
182 stars 43 forks source link

FloatingPointError: Predicted boxes or scores contain Inf/NaN. Training has diverged. #37

Open xiaohei1001 opened 2 years ago

xiaohei1001 commented 2 years ago

iter: 1 total_loss: 4.02e+06 loss_cls: 3.218e+06 loss_box_reg: 4.646e+05 loss_rpn_cls: 1.197e+05 loss_rpn_loc: 2.168e+05 data_time: 9.5348 lr: 2e-05 max_mem: 5626M FloatingPointError: Predicted boxes or scores contain Inf/NaN. Training has diverged.

I experimented on COCO dataset with default settings，and the above error occurred.

er-muyue commented 2 years ago

Hi, same with this issue. Please use detectron0.3，the default version is 0.5

xiaohei1001 commented 2 years ago

Hi, same with this issue. Please use detectron0.3，the default version is 0.5

@er-muyue Thank you！ Why do higher versions of Detectron have errors?

BeverYang commented 1 year ago

更高版本的detectron2应当只导致参数错误，不应导致训练爆炸

ljx0114 commented 3 months ago

After changing to 0.3, the problem still exists

dashenfeng commented 2 weeks ago

After changing to 0.3, the problem still exists

I found the same problem, do you solve it?

ljx0114 commented 2 weeks ago

No, I gave up.

 

------------------ 原始邮件 ------------------ 发件人: "er-muyue/DeFRCN" @.>; 发送时间: 2024年7月9日(星期二) 上午8:35 @.>; @.**@.>; 主题: Re: [er-muyue/DeFRCN] FloatingPointError: Predicted boxes or scores contain Inf/NaN. Training has diverged. (Issue #37)

After changing to 0.3, the problem still exists

I found the same problem, do you solve it?

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