Creating mask Datastore for Mask R-CNN

I am working with 126 (size 512x512) grayscale images. Each image has atleast 10-100 object instances. I labeled polygon masks and bounding boxes with the use of the image labeler app. I exported the ground truth and separated the images and bbox in their own datastore. Now I am looking to create the imagedatastore for the masks by following the instructions on:

https://www.mathworks.com/help/vision/ug/getting-started-with-mask-r-cnn-for-instance-segmentation.html

Create Datastore that Reads Data: Create an imageDatastore and specify a custom read function that returns mask data as a binary matrix

I went on to follow the steps of converting the polygons labels to maskstacks. The mask stacks I got were as expexted the mask (h,w,numobjects). For example one of my maststacks was a 512x512x39 logical .

How do I format these mask stacks into an image datastore to train the mask RCNN.

Answer 1

[T.Nikhil kumar]

Hi Bryce,

According to the documentation you’re following, the next step for you is to create an ‘imageDatastore’ that reads the mask data as a binary matrix.

I would suggest you to try this possible solution for the same.

1. After converting polygon labels into binary masks, you can save each mask stack in a .mat file with a suitable name (For eg: maskStack1.mat for the first image). You can put all these individual mask stack files in a ‘MaskStackFolder’ folder.

2. Now, define a custom read function that should take a file path as input and return the binary mask stack in the expected format. Refer to the following code snippet:

function mask = customReadFcn(filename)
    % Load the data from the .mat file
    loadedData = load(filename);
    % Assuming the variable containing the mask stack is named 'maskStack'
    mask = loadedData.maskStack;
    % Ensure you are using correct variable name as per your naming
end

3. You can now create an ‘imageDatastore’ with this folder and custom read function.

maskFolder = "C:\MATLAB\MaskStackFolder"; % Replace with the actual path to your mask stack folder

% Create an imageDatastore for the mask stacks
masksDatastore = imageDatastore(maskFolder, 'ReadFcn', @customReadFcn);

Ensure that the order of the mask stacks in the ‘imageDatastore’ corresponds to the actual order of the images in the previously create image datastore for images.

I hope this answer is able to help you!

[Bryce]

Hey T.Nikhil kumar,

Thanks for your answer. This problem has been a road block in my work so I am thankful for your answer. Unfortunately I tried the option and got this error.

Error using imageDatastore

Input folders or files contain non-standard file extensions.

Use FileExtensions Name-Value pair to include the non-standard file extensions.

I tried using the input argumet for imagedatastore ("FileExtenstions",".mat") to attempt to resolve the error and it worked so far. Hopefully I don't run into any other problems :). I am still curious why they did not put this in the instuctions/website