How should I test the trained model?

[QiaoJinming]

I have successfully trained the model, but I don't know how to test the model,may i catch modified code，thank you very much

[zhijiejia]

Here is the test on the test dataset

[QiaoJinming]

Here is the test on the test dataset

Thank you very much for your reply, IoU is satisfactory, but when I use two images (refer and test) to predict the mask (label), I get very bad results. According to your code, I use it like this "output"

res = output.squeeze(0).cpu().numpy() plt.imshow(res)

Got this result

The result is very frustrating, but the code you designed is very effective. I suspect that the method I used is wrong, so I want to ask you to get a copy of the execution code for generating the label from the refer and test images. If you interrupt, please understand.

Thanks again for your work

[QiaoJinming]

Here is the test on the test dataset

Supplement the reply just now

Hyperparameters have not been modified
hardware
nvidia RTX 2080
software
torch==1.7.0
PIL==8.0.1
python==3.7

Thank you

[QiaoJinming]

Hello, I have replied to your answer, but I have not received a new notification. Do you have time to reply? The network architecture of this project is very influential. I sincerely look forward to your reply.

[zhijiejia]

All my code is here, and the effect of this network itself is not very good. I'm going to test the code again later。

[sssimpleboy]

I have successfully trained the model, but I don't know how to test the model,may i catch modified code，thank you very much我已经成功训练了模型，但我不知道如何测试模型，我可以捕获修改后的代码，非常感谢

您好 请问您知道如何测试模型了吗？ 我的ouput 是 2，224，224 我不知道怎么转化成 224，224 的黑白图 哈哈哈哈