Deep Learning Model(CNN) Question on pynq

Hi, I have seen many papers about using pynq to accelerate AI model such as CNN.
I want to create my own CNN model using pynq overlay to achieve accleration.
I have download Vitis AI to run some example with its overlays.
These are the problems I am confused about:
(1) Can I retrain these model such as yolo included in VITIS AI with my own dataset?
(2) Even more, can I build my own model using CNN with overlay on the board using tools like REVISION or other tools?
The board I use is ultra96 v2.
Any advice will be appreciate!
Thanks for your helps!
Tim

From my knowledge, ultra96 v2 is quite different from the pynq family.

pynq uses the Zynq FPGAs whereas ultra96 uses the ultrascale FPGAs.

Have you read the entire Vitis AI documentation on the Xilinx DocNav?