cannot generate exactly the same projection maps from original 2D slices

[ustlsh]

Hi,

I am wondering about the process of generating projection maps B1 and B4. In the paper, it says that B1 and B4 are the average projection of OCT and OCTA volumes accordingly. I tried to use the original b-scan images, i.e., .bmp files to generate them. But I cannot get exactly the same value of the downloaded projection maps.

What I have done is, for 3M images:

1. Concatenate 304 2D slices (304x640) along dim0, to form 304x304x640 volume.
2. Use img = np.mean(volume, axis=0) to project to 304x304 size image.
3. Normalize to 0-255 by (img-img.min)*255/(img.max-img.min)
4. Flip img to align with the given projection labels.
   The results look darker than the given projection maps. I also check the value for each pixel, and they are different. Any process that I miss or anywhere I make the mistake?
   Projected by above steps:

Ground truth projection map:

[chaosallen]

Some singular values need to be removed before normalization. There, we give the matlab code: x=prctile(Proj1(:),99); Proj1(Proj1>x)=x; Proj1=(Proj1-min(min(Proj1)))/(max(max(Proj1))-min(min(Proj1)));

…

------------------ 原始邮件 ------------------ 发件人: "chaosallen/IPNV2_pytorch" ***@***.***>; 发送时间: 2022年1月4日(星期二) 中午1:27 ***@***.***>; ***@***.***>; 主题: [chaosallen/IPNV2_pytorch] cannot generate exactly the same projection maps from original 2D slices (Issue #6) Hi, I am wondering about the process of generating projection maps B1 and B4. In the paper, it says that B1 and B4 are the average projection of OCT and OCTA volumes accordingly. I tried to use the original b-scan images, i.e., .bmp files to generate them. But I cannot get exactly the same value of the downloaded projection maps. What I have done is, for 3M images: Concatenate 304 2D slices (304x640) along dim0, to form 304x304x640 volume. Use img = np.mean(volume, axis=0) to project to 304x304 size image. Normalize to 0-255 by (img-img.min)*255/(img.max-img.min) Flip img to align with the given projection labels. The results look darker than the given projection maps. I also check the value for each pixel, and they are different. Any process that I miss or anywhere I make the mistake? Projected by above steps: Ground truth projection map: — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>