Why do testnet.py test results (AP,AP50,AP75,APm,APs,APl) all go to 0?

[906096481]

For example ('AP',2.2241e-05)

[krumo]

Hi, please consider to provide more details about your experimental setting (e.g. your PyTorch version, your cuda version, the task you are working on and so on). If possible, you are welcome to provide your training log. Only with more details could I give you more help.

[906096481]

Hi,This is my training log.The environment is configured as required on GitHub

…

------------------ 原始邮件 ------------------ 发件人: "krumo/Domain-Adaptive-Faster-RCNN-PyTorch" ***@***.***>; 发送时间: 2022年4月18日(星期一) 晚上6:14 ***@***.***>; ***@***.******@***.***>; 主题: Re: [krumo/Domain-Adaptive-Faster-RCNN-PyTorch] Why do testnet.py test results (AP,AP50,AP75,APm,APs,APl) all go to 0? (Issue #38) Hi, please consider to provide more details about your experimental setting (e.g. your PyTorch version, your cuda version, the task you are working on and so on). If possible, you are welcome to provide your training log. Only with more details could I give you more help. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[krumo]

Unfortunately I cannot find where you place your training log. Could you upload it again?

[906096481]

log.txt

[906096481]

I realized that maybe the weights were not loaded during the test, so the AP value tended to be 0, but I did write it in.

[krumo]

In the training log, we could observe that after training the final model is tested on Foggy Cityscapes val and a performance of ~39.8 mAP@50 is obtained. Did you follow the README.md to run
python tools/test_net.py --config-file "configs/da_faster_rcnn/e2e_da_faster_rcnn_R_50_C4_cityscapes_to_foggy_cityscapes.yaml" MODEL.WEIGHT <path_to_store_weight>/model_final.pth to test your final model?

[906096481]

Yes, I followed the tutorial. I know the training results are fine, but the test results don't know why.

[906096481]

I tried writing the weight path directly to default.py, and the result was the same.

[krumo]

It seems the training process is okay. Could you also upload your test log here?

[906096481]

[krumo]

This screenshot doesn't provide any meaningful information. Please make sure you load the correct checkpoint when performing tests.

[906096481]

This screenshot doesn't provide any meaningful information. Please make sure you load the correct checkpoint when performing tests.

I have filled in the weight address here, is it correct?

[906096481]

I have the same problem,maybe we can contact each other to solve the problem in a convenient way such as wechat.

ok.

[906096481]

I have the same problem,maybe we can contact each other to solve the problem in a convenient way such as wechat.

ok.

wechat number:18205202287

OK

[zhengying455]

I have the same problem,maybe we can contact each other to solve the problem in a convenient way such as wechat.

ok.

wechat number:18205202287

OK

打扰您了，我最近在学习这个方向相关的知识，在复现代码的是时候遇到了一些问题，或许可以加一下您的微信请教一下吗，如果可以的话，非常感谢