Mean class IoU of photo2label

[mxkai]

Hello,can you tell me how to calculate Mean class IoU, Mean class accuracy and Mean pixel accuracy of photo2label

[junyanz]

See the function get_score here.

[mxkai]

I want to calculate the scores of the original label，but I meet some difficulties. I use the script of cityscapes to generate file "_gtFine_labelTrainIds.png", and then replace the FCN output with these files： but the output is bad：

[ZhangCZhen]

Hello, I evaluated the results after using the cityscape dataset, and the values obtained are far from those mentioned in the paper. So I would like to ask, is the code given by 'evaluate.py' only used to evaluate the results of label2image? Can I evaluate image2label? After I use RGB image to generate semantic segmentation results, how do I evaluate the segmentation results? Thanks a lot!

[mxkai]

I want to train photo2label on cityscapes，Can you tell me whether the training label is "_gtFine_color.png" or "_gtFine_labelTrainIds.png"？ _gtFine_color.png： _gtFine_labelTrainIds.png：

[ZhangCZhen]

I use the "_gtFine_color.png", but I don't know if it is possible to evaluate the results of semantic segmentation with the Fcn model.

[mxkai]

Is the semantic color of the generated image random? Whether it can correspond to the label？

[ZhangCZhen]

I think it depends on your generator performance.

[ZhangCZhen]

I have the same question. Sorry.

[ZhangCZhen]

@mxkai Hello.

1. My experimental results are the same as yours, they are 0.997, 0.209, 0.2058. If you uncomment this line of code 'scipy.misc.imresize(im, (256,256))' , the value will become 0.93, 0.597, 0.452. I think it should be this reason.
2. I think the training label should be '_gtFine_color.png'.
3. Do you know how to convert those color label into gray label (or TrainId) easily? Thanks a lot. Looking forward to your reply.

[A-Little-Nut]

@ZhangCZhen
Have you solved the problem how to evaluate on cityscapes? I meet the same problem. Can you help me? How should I change the original code?

[ZhangCZhen]

@ZhangCZhen Have you solved the problem how to evaluate on cityscapes? I meet the same problem. Can you help me? How should I change the original code?

I have evaluate 'label2photo'. I am also confused how to evaluate 'photo2label'.

[junyanz]

See this post for more details regarding photo2label evaluation.

[HuXiaokai12138]

I want to calculate the scores of the original label，but I meet some difficulties. I use the script of cityscapes to generate file "_gtFine_labelTrainIds.png", and then replace the FCN output with these files： but the output is bad：

@mxkai I would like to ask if you solved the problem of evaluating photo2label? I'm also having the same issue and look forward to your reply.