The client must now create a dataset where they have enough images containing the concept they wish to detect/segment. The building of this dataset totally depends on their task, what they are trying to achieve, and what their end goal is. Regardless, the following steps must be followed properly:

• The images or videos they capture must be as close to the deployment settings as possible. If they are using GoPro images in deployment, then they should use those in building the dataset too. If they are crowd-sourcing images for deployment, then they should do that here too