Designing an interactive decision tree for vision model deployments

For many popular deployment devices, such as the NVIDIA Jetson edge computer, there are specific intricacies that only apply to that device, and that may vary depending on the version of operating system you are using. To add to the complexity, many combinations of models and deployment devices are not supported. You cannot deploy CogVLM on a Raspberry Pi due to the size of the model and GPU requirements, but you can deploy it on an NVIDIA T4.

Clearly communicating how to deploy a computer vision model, and ensuring we clearly enumerate which combinations of devices and models are supported, is essential for the Roboflow team. The less friction a customer runs into when deploying a model, the faster they can get to production.

The status quo was a series of documentation pages, many external and some internal, that we used to communicate deploying models. But, we wondered if we could do better. Could we offer a more interactive experience to users who were deploying a model for the first time? Herein came the idea for a single page where a user could get everything they need to deploy a model.

In this blog post, I walk through how we went from concept to an executed product that is available for customers to use and actively linked in relevant documentation. Here is a video showing the system in action:

Leave a Comment