Chasel-Tsui / mmdet-rfla

ECCV22: RFLA
MIT License
247 stars 23 forks source link

老师您好,我使用您公开的官方代码跑AI-TOD,结果跟您论文里的差很多,请问是怎么回事呀 #58

Open ghost opened 10 months ago

Chasel-Tsui commented 10 months ago

您好,请问可以提供更多细节吗,比如使用的模型是什么,显卡,batch size,learning rate等等

ghost commented 10 months ago

老师您好,感谢您百忙之中回复,使用的是faster rcnn r50,显卡是A6000,batchsize与您论文一致是2,learning rate是默认的1x,其他配置都与您开源的相同,没有修改

---- 回复的原邮件 ---- | 发件人 | Chang @.> | | 日期 | 2023年09月16日 04:04 | | 收件人 | @.> | | 抄送至 | @.>@.> | | 主题 | Re: [Chasel-Tsui/mmdet-rfla] 老师您好,我使用您公开的官方代码跑AI-TOD,结果跟您论文里的差很多,请问是怎么回事呀 (Issue #58) |

您好,请问可以提供更多细节吗,比如使用的模型是什么,显卡,batch size,learning rate等等

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

Chasel-Tsui commented 10 months ago

可以把训练使用的config和第一个epoch的log以及最后的结果发我看一下吗,有可能在训练的初期模型没有收敛

ghost commented 10 months ago

您好,这是配置文件以及训练的log文件

在 2023-09-17 16:52:28,"Chang Xu" @.***> 写道:

可以把训练使用的config和第一个epoch的log以及最后的结果发我看一下吗,有可能在训练的初期模型没有收敛

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

Chasel-Tsui commented 10 months ago

您好,不好意思我这边收不到邮件和附件,可以直接在github issue上面贴一下文字或者截图吗