Why did I get a loss of 5e-5 after running "train_uv_modify.py", but the simulation result was particularly poor? In order to get the same result as" NS_model_train.pt", do I need to pay attention to anything else ? Thank you very much!

[uigytkilod]

No description provided.

[Shengfeng233]

the loss include both data loss and equation loss, if both the data loss and equation loss are declined around 1e-5, the flow field should be well fitted (according to my experiences and the recent literature). So I suggest you may check if there was something wrong with the post processing (plotting). I think the only thing that is so tricky is the loss function, and as soon as the loss can be effectively declined, there are nothing more to pay attention to.