I able to apply DeepLabV3+ to segment the images, but also like to get the boundary around individual detection.
For example, in the image segmentation mask above, I cannot distinguish between the two children on the horse. If I could draw the boundary around each individual children or put a different color for them, I would be able to distinguish them. Please let me know if is there any way to configure deepLab to achieve that.
You are confusing two tasks: semantic segmentation and instance segmentation.
DeepLbV3+ (and many similar deep nets) are solving semantic segmentation problem: that is labeling each pixel with the class it belongs to. You got a very nice results where all pixels belonging to "person" were colored pink. Semantic segmentation algorithms do not care how many "person"s there are in the image and they do not wish and do not care to label each person separately. As long as all "person" pixels were labeled as such - the task is considred well done.
On the other hand, what you are looking for is instance segmentation: that is labeling each "person" as a unique person in the image. This is far more complex task: not only should you succeed in labeling all "person" pixels as "person", but also you want to group the "person" pixels into the different instances in the image.
Since instance segmentation is a more difficult task, you would need different models/nets to accomplish it.
I suggest Mask R-CNN as a good starting point for instance segmentation algorithms.