Linfeng-Tang / SwinFusion

This is official Pytorch implementation of "SwinFusion: Cross-domain Long-range Learning for General Image Fusion via Swin Transformer"
161 stars 20 forks source link

模型训练中的损失函数收敛问题 #20

Open jiqingtao123 opened 10 months ago

jiqingtao123 commented 10 months ago

作者 您好! 我有一个小问题 源代码中的train.log中的G_loss 总损失好像没有收敛 那近红外和可见光图像的模型训练完成是以什么指标作为标准的呢 希望能够指导一下 谢谢

Linfeng-Tang commented 10 months ago

对于图像融合任务而言 会存在loss可能没有收敛 但是融合的需求能够实现的情况。一般需要根据经验来设置训练的迭代数。

jiqingtao123 @.***> 于2023年9月12日周二 21:19写道:

作者 您好! 我有一个小问题 源代码中的train.log中的G_loss 总损失好像没有收敛 那近红外和可见光图像的模型训练完成是以什么指标作为标准的呢 希望能够指导一下 谢谢

— Reply to this email directly, view it on GitHub https://github.com/Linfeng-Tang/SwinFusion/issues/20, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOFH5ILY3VGERBFGQM4PV7LX2BOMHANCNFSM6AAAAAA4U3D3AI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

jiqingtao123 commented 10 months ago

太感谢了

m77-5 commented 2 months ago

请问用的是什么版本的python呢,我这边运行总是有库不匹配的情况

Linfeng-Tang commented 2 months ago

我本地使用的是Python3.8版本

m77-5 @.***> 于2024年4月17日周三 17:34写道:

请问用的是什么版本的python呢,我这边运行总是有库不匹配的情况

— Reply to this email directly, view it on GitHub https://github.com/Linfeng-Tang/SwinFusion/issues/20#issuecomment-2060829362, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOFH5IOQLBBNHZUOVUP2BRLY5Y625AVCNFSM6AAAAAA4U3D3AKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANRQHAZDSMZWGI . You are receiving this because you commented.Message ID: @.***>