Open watchmexiang opened 5 years ago
We did not use MIoU as a measurement to evaluate the methods. When you use FCN, what is the output? The output layer is 'softmax' or 'sigmoid' ? In our implementation, the output of FCN is 'sigmoid'. When you compute the MIoU, did you take the background into account? In our case, the background should not be considered. In addition, instead of using MIoU, we use AIU (average intersection over union) as a metric to evaluate the compared methods. For the specific definition of AIU, you are referred to the paper.
For the code to compute ODS and OIS, you are referred to https://github.com/s9xie/hed. BTW, if you use softmax output, the predicted crack map is a binary map, but sigmoid gives you a probability map. This may cause some problems when using MioU metric. For example, how to binarize the probability map? what kind of threshold should be used to fairly evaluate the predicted results.
Fan
On Wed, May 15, 2019 at 6:03 AM zhangyuqiqiyuzhang notifications@github.com wrote:
Or can you share the following code for calculating both ODS and OIS indicators?
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/fyangneil/pavement-crack-detection/issues/4?email_source=notifications&email_token=ALAUWTXBEVGY5EANDFFG3E3PVPNYFA5CNFSM4HM3EMEKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODVOFRGI#issuecomment-492591257, or mute the thread https://github.com/notifications/unsubscribe-auth/ALAUWTQXCOZGG6MP4TPUDALPVPNYFANCNFSM4HM3EMEA .
@zhangyuqiqiyuzhang The evaluation tool is provided, please check it out.
Thank you for your continued help, can you share the following experimental results of crackforest in each data set, thank you again for your help.
Or you can send it to my email address, 1264754202@qq.com, thanks again
I am really sorry to bother you again. I tried to use CrackForest to run some data sets you used to get the test results, but I can't get your result map. It may be my operation problem, so if you are convenient, can you Share the results of Crackforest's experiments in the dataset. If it is not convenient, can you give me some help? Thank you again for your help
i am a little busy recently. I will share you the result in this weekend, if i have it.
Fan
On Thu, Jun 6, 2019 at 9:53 PM zhangyuqiqiyuzhang notifications@github.com wrote:
I am really sorry to bother you again. I tried to use CrackForest to run some data sets you used to get the test results, but I can't get your result map. It may be my operation problem, so if you are convenient, can you Share the results of Crackforest's experiments in the dataset. If it is not convenient, can you give me some help? Thank you again for your help
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/fyangneil/pavement-crack-detection/issues/4?email_source=notifications&email_token=ALAUWTSRMLB22JQQWLAAU4DPZHSTBA5CNFSM4HM3EMEKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXE2WQA#issuecomment-499755840, or mute the thread https://github.com/notifications/unsubscribe-auth/ALAUWTWPGKGJUDRWTH4RPHTPZHSTBANCNFSM4HM3EMEA .
@zhangyuqiqiyuzhang I have uploaded it to google drive, please check it out.
I have successfully downloaded the test results, thank you again for your help.
I'd like to ask you the difference between traincrop and traindata in the Crack500,thanks!
the image in train data is raw image. Images in traincrop are cropped patches on raw image
Get Outlook for iOShttps://aka.ms/o0ukef
From: fightingchun notifications@github.com Sent: Thursday, November 26, 2020 8:32:29 AM To: fyangneil/pavement-crack-detection pavement-crack-detection@noreply.github.com Cc: Fan Yang fyang@temple.edu; Comment comment@noreply.github.com Subject: [External] Re: [fyangneil/pavement-crack-detection] problem (#4)
I'd like to ask you the difference between traincrop and traindata in the Crack500,thanks!
— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/fyangneil/pavement-crack-detection/issues/4#issuecomment-734300503, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ALAUWTUCNDQNGVXJM7NM673SRZKG3ANCNFSM4HM3EMEA.
Hello, I used the original FCN to experiment on your dataset and found that the results obtained by FCN are better than the results of your model under the MIOU indicator. After reading your article, I found that you used The indicator is the indicator of edge detection. You have come to the conclusion that your model is better than FCN. Have you compared the test results of the two models under the MIOU indicator? Thank you.