search

Minerva-J / Pytorch-Segmentation-multi-models

Pytorch implementation for Semantic Segmentation with multi models (Deeplabv3, Deeplabv3_plus, PSPNet, UNet, UNet_AutoEncoder, UNet_nested, R2AttUNet, AttentionUNet, RecurrentUNet,, SEGNet, CENet, DsenseASPP, RefineNet, RDFNet)
Apache License 2.0
179 stars 43 forks source link

How to test the model results and accuracy? #1

Open GinkgoX opened 3 years ago

GinkgoX commented 3 years ago

Dear Minerva-J: Very nice work for DRIVE datatset segmentation. could you please share the test.py and give the accuracy metrics calcualtion method? thanks!

GinkgoX commented 3 years ago

sorry for that, could you have a brief introduction of the dataset architecture and processing?

Minerva-J commented 3 years ago

Dear Minerva-J: Very nice work for DRIVE datatset segmentation. could you please share the test.py and give the accuracy metrics calcualtion method? thanks! Thank you, the test phase is also in train.py.

Minerva-J commented 3 years ago

sorry for that, could you have a brief introduction of the dataset architecture and processing?

Hello,The DRIVE dataset is available at https://www.isi.uu.nl/Research/Databases/DRIVE/.

GinkgoX commented 3 years ago

I see the test method, so would you mind to share with the method with multi-class segmentation method? I find no visualized image generated, and strange this is that the dice result is almost 0 while the accuracy is above 90%, Is there anything wrong? thanks a lot!

GinkgoX commented 3 years ago

Dear Minerva-J: In your code, you the multi-class segmentation method on DRIVE dataset, would you mind share with the labeled mask(namely the whole dataset for trianing and test) or your private dataset link, I still meet the above BUG.

Minerva-J commented 3 years ago

I see the test method, so would you mind to share with the method with multi-class segmentation method? I find no visualized image generated, and strange this is that the dice result is almost 0 while the accuracy is above 90%, Is there anything wrong? thanks a lot!

you should modify class num in the code for multi-class segmentation.

GinkgoX commented 3 years ago

I see the test method, so would you mind to share with the method with multi-class segmentation method? I find no visualized image generated, and strange this is that the dice result is almost 0 while the accuracy is above 90%, Is there anything wrong? thanks a lot!

you should modify class num in the code for multi-class segmentation.

Dear Minaerva-J: as your response, I have change the class num and the multi-class label as 0, 1, 2, 3, 4 for single channel mask, the training method is Ok, while there is no visulized result in the result folder. Could you please help me about this problem, thanks a lot lot lot !

xiaoyuamw commented 3 years ago

Dear Minerva-J: Very nice work for DRIVE datatset segmentation. could you please share the test.py and give the accuracy metrics calcualtion method? thanks!

Hi,@GinkgoX Due to my limited ability, i still can't found how to test one of your project, such as deeplabv3, can you tell me how to do it?

Best Regards.

PatrickWilliams44 commented 2 years ago

Dear Minerva-J: It's my pleasure to see your segmentation-multi-models! Now, i have some troubles about these models. Can you tell me where is the "train_test_loader" of the test.py in the train.py! Looking forward to your early reply!

Minerva-J commented 2 years ago

I see the test method, so would you mind to share with the method with multi-class segmentation method? I find no visualized image generated, and strange this is that the dice result is almost 0 while the accuracy is above 90%, Is there anything wrong? thanks a lot!

you should modify class num in the code for multi-class segmentation.

Dear Minaerva-J: as your response, I have change the class num and the multi-class label as 0, 1, 2, 3, 4 for single channel mask, the training method is Ok, while there is no visulized result in the result folder. Could you please help me about this problem, thanks a lot lot lot !

Hi, visulize for colorize_mask in train.py just suitable for one channel mask or RGB mask, sorry about that.

Minerva-J commented 2 years ago

Dear Minerva-J: Very nice work for DRIVE datatset segmentation. could you please share the test.py and give the accuracy metrics calcualtion method? thanks!

Hi,@GinkgoX Due to my limited ability, i still can't found how to test one of your project, such as deeplabv3, can you tell me how to do it?

Best Regards.

Hi, you just modify train.py to coding test.py, img loader and model load is the same ,then just run test fuction to test your data, and --model deeplabv3

Minerva-J commented 2 years ago

Dear Minerva-J: It's my pleasure to see your segmentation-multi-models! Now, i have some troubles about these models. Can you tell me where is the "train_test_loader" of the test.py in the train.py! Looking forward to your early reply!

ok, it is my fault, the train_test_loader should be correct to test_loader, which is just same as val_loader.