Questions about using custom images as learning data in the Yolov4 example model

I tried to learn using custum images in the yolov4 example model provided by mathworks.

The problem is how to process learning images.

An image is the content of the dataset used in the example. Here, it is determined that the value displayed on the 'vehicle' is not the content of the boundary box position. And the information in the bounding box was created somewhere when creating the image datastore later, but I don't know the principle.

What I tried was to load the mat file labeled after labeling through imageLabeler and proceed with the example, but the detector did not work properly after learning.

What process should I try?

thanks.

Answer 1

[Sanjana]

Hi,

I understand that you are facing an issue with understanding how the “bounding box” data is generated for the objects in the images using “Image Labeler” application in MATLAB.

As per the example, the values specified under “vehicle” are the “top left corner, width and height” values for the bounding boxes.

When using “Image Labeler” app in MATLAB, after you label the “vehicles” in the Custom dataset images and export the labelled data as a “groundTruth” object, you can find the “bounding box” labels in the “LabelData” property of the generated “groundTruth” object.

Please refer to the following link, for further information,

https://www.mathworks.com/help/vision/ug/training-data-for-object-detection-and-semantic-segmentation.html

Hope this helps!

Regards,

Sanjana