Application requirements can be categorized into three categories:
- Using the models that are provided by the AMD Vitis™ AI Library to build your own application.
- Using your own custom models which are similar to the models in the Vitis AI Library.
- Using new models that are totally different from the models in the Vitis AI Library.
This chapter describes the development steps for the first two cases. For the third case, you can use the Vitis AI Library samples and implementation for reference. This chapter provides information on:
- Customizing preprocessing
- Using the configuration file to specify the preprocessing and postprocessing parameters
- Using the Vitis AI Library's postprocessing library
- Implementing user postprocessing code
- Working with the xdputil tool
The following figure shows the relationship between the Vitis AI Library APIs and their corresponding examples. There are four kinds of APIs in this release:
- Vitis AI Library API_0 based on VART
- Vitis AI Library API_1 based on AI Library
- Vitis AI Library API_2 based on DpuTask
- Vitis AI Library API_3 based on Graph_runner
Figure 1. The Diagram of AI Library API

Choosing an API for Your Application
Use the following recommendations to choose an API for your application.
- If the model has been split into several subgraphs, API_3
Graph_runner
is recommended for model deployment. - If the model has a custom op, API_3
Graph_runner
is recommended for model deployment. - If you want to get the best performance and you are a beginner
at using AI algorithms, such as model, preprocessing, and postprocessing, API_1
AI_Library
is recommended. - If you want to use AMD models to quickly
build applications, API_1
AI_Library
is recommended. - If you have your own models that are retrained using your own
data under the Vitis AI library support network list, API_1
AI_Library
is recommended. - If you want to use your custom preprocessing or postprocessing
algorithms, API_2
DpuTask
is recommended. - If you want to develop and apply AI algorithms on multiple
platforms and you are an advanced user of AI algorithms, API_0
VART
is recommended.