Clear Filters
Clear Filters

How to interpret the wrong image background of data augmentation?

35 views (last 30 days)
I'm working in semantic segmentation and faced a problem with the output of data augmentation after running the attached code. This problem is related to the background of some ground truth images; instead of a black background, it displays white, as shown in the attached screenshot.
for j = 1:numImages
% Read the original image and ground truth
img = readimage(imdsClass, j);
gtImg = readimage(gtImdsClass, j);
gtImg=imcomplement(gtImg);
% Generate filenames for the original images
originalImgName = fullfile(classImageFolder, ['original_' num2str(j) '_' classLabel '.jpg']);
originalGtImgName = fullfile(classGTFolder, ['original_' num2str(j) '_' classLabel '.png']);
% Save the original image and ground truth
imwrite(img, originalImgName);
imwrite(gtImg, originalGtImgName);
end
% Perform augmentation for the images that need to be generated
for j = 1:numToGenerate
% Read an image and its corresponding ground truth
img = readimage(imdsClass, mod(j, numImages) + 1);
gtImg = readimage(gtImdsClass, mod(j, numImages) + 1);
gtImg=imcomplement(gtImg);
% Apply augmentation (random rotation and flipping)
x = randi([-30, 30], 1); % Random rotation angle
y = randi([1, 2], 1); % Random flip direction (1: vertical, 2: horizontal)
% Augment the image and ground truth
augmentedImg = imrotate(img, x, 'bilinear', 'crop');
augmentedImg = flip(augmentedImg, y);
augmentedGtImg = imrotate(gtImg, x, 'nearest', 'crop');
augmentedGtImg = flip(augmentedGtImg, y);
% Generate filenames for the augmented images
augmentedImgName = fullfile(classImageFolder, ['augmented_' num2str(j) '_' classLabel '.jpg']);
augmentedGtImgName = fullfile(classGTFolder, ['augmented_' num2str(j) '_' classLabel '.png']);
% Save the augmented image and ground truth
imwrite(augmentedImg, augmentedImgName);
imwrite(augmentedGtImg, augmentedGtImgName);
end
  2 Comments
Shivansh
Shivansh on 8 Sep 2024 at 10:24
Hi Abdulrahman,
I tried to reproduce the issue but the code works fine on my end.
Please share a few images for reproducing the issue.

Sign in to comment.

Answers (1)

Image Analyst
Image Analyst on 8 Sep 2024 at 14:39
Why are you calling imcomplement? Try not doing that.
What is the original background: white or black?
Finally, you could just check if it's white in the upper left pixel, and invert it if it's white.
if augmentedImg(1,1) ~= 0
augmentedImg = ~augmentedImg; % or imcomplement(augmentedImg), whatever works.
end
  2 Comments
Abdulrahman
Abdulrahman on 8 Sep 2024 at 23:01
The original background images is black.
My problem is how to unifirm the background of all data as shown in attached pic.
I run the code without gtImg=imcomplement(gtImg);for small data (attached DA) and got correct results using your code (if augmentedImg(1,1) ~= 0 ......) but I got wrong results for whole dataset.
Image Analyst
Image Analyst on 9 Sep 2024 at 1:33
Edited: Image Analyst on 9 Sep 2024 at 1:33
Why are you manually augmenting your images instead of using augment?

Sign in to comment.

Categories

Find more on Image Data Workflows in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!