search

emma-sjwang / pOSAL

Code for TMI paper: Patch-based Output Space Adversarial Learning for Joint Optic Disc and Cup Segmentation
MIT License
51 stars 19 forks source link

I have a problem of the dataset. #2

Closed zhaoyihao12345 closed 4 years ago

zhaoyihao12345 commented 4 years ago

I found that not all of the OD regions are covered by the cropped ROI in Drishti-GS dataset provided by you. Why?

emma-sjwang commented 4 years ago

Could you provide more information? Actually, In my previous experiments, the OD regions can successfully be covered! If not, you can adjust the radius to make it be covered.

zhaoyihao12345 commented 4 years ago

I am very glad to hear from you. The OD regions are not completely covered by the cropped ROI, for example, the images named 'gdrishtiGS_032.png' and 'gdrishtiGS_005.png' in Drishti-GS dataset provided by you.

emma-sjwang commented 4 years ago

I am sorry that I did not find the 'gdrishtiGS_032.png' in the results link that I provided. For the figure 'gdrishtiGS_005.png', it is well segmented which means the ROI is cropped correctly. Please download the results from the project website https://emmaw8.github.io/pOSAL/index.html.

zhaoyihao12345 commented 4 years ago

The two images are 'data/Drishti-GS/train/disc_small/image/gdrishtiGS_032.png' and 'data/Drishti-GS/test/disc_small/image/gdrishtiGS_005.png'. I can find them in your github.

------------------ 原始邮件 ------------------ 发件人: "Shujun WANG"<notifications@github.com>; 发送时间: 2019年11月18日(星期一) 下午4:40 收件人: "EmmaW8/pOSAL"<pOSAL@noreply.github.com>; 抄送: "1152048717"<1152048717@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [EmmaW8/pOSAL] I have a problem of the dataset. (#2)

I am sorry that I did not find the 'gdrishtiGS_032.png' in the results link that I provided. For the figure 'gdrishtiGS_005.png', it is well segmented which means the ROI is cropped correctly. Please download the results from the project website https://emmaw8.github.io/pOSAL/index.html.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

emma-sjwang commented 4 years ago

If you cannot accept it, you can change the cover range to cover the ROI. I uploaded an optional training set named 'disc_small' for your reference you can design your own dataset according to a reasonable range. In my experiment, I may use the ROIs saved in 'disc' but I did not release it. For the test ROI, actually it didn't matter since we directly predict the results from the whole image, the cropped ROI was used for evaluation during the training. Hope this can help you.

zhaoyihao12345 commented 4 years ago

Thank you very much.