Search code examples
machine-learningconv-neural-networkobject-detection

How to get the outline of the image detected (using cnn object detection) rather than rectangular box around it?


I have been googling for 1 week about object detection. What everyone show is similar to the below image. enter image description here

But I don't want square boxes around the detected object. I need the rough outline of the object as shown in the below image. enter image description here

Is it possible? If possible please give a brief explanation about it. Thanks in advance.


Solution

  • I think you are looking for semantic segmentation. Object detection usually means finding a bounding box and a label for an object, while semantic segmentation is the problem of assigning a class label to each pixel in an image. Once you have the class labels for each pixel, you can use a classical contour detection algorithm to find the outline of the object. Matlab docs have a nice explainer on semantic segmentation.

    Below is an example output of a semantic segmentation model trained on a self-driving car dataset:

    enter image description here

    Image by Vladlen Koltun