Error Inferring Yolov8 node_args.cpp Vitis Ai provider

[farah-rostom98]

Describe the issue

Hello,

I had a quantized Yolov8 model that I wanted to infer on a Kria-KV260 target with DPUCZDX8G-B409 with Vitis AI option. During quantization, I excluded the nodes that aren't supported by the DPU architecture like: reshape, resize, slice, split, divide, and subtract. The model runs with onnx run time successfully on my local laptop on CPU. However, when I tried to run it on the target, there was an error -> node_arg.cpp:329, unknown type:2, check failure stack trace, Aborted. I changed the logging to be verbose so that I could have more info and I am attaching the output of the log. I tried to look what would be inside node_arg.cpp file to debug but couldn't find any clues.

Would appreciate if you can help debugging this. I have attached the quantized model and inference script as well for your reference.

To reproduce

1- Move Quantized model to the kit
2 - Used the following script to create a session

Yolov8.zip

Urgency

I am currently working on a project trying to assess the performance of FPGAs vs GPUs and this issue is blocking me as this task aim was to estimate how can we scale the FPGA vs the GPU.

Platform

Linux

OS Version

Peta-Linux, Vitis3.0

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

1.14.0

ONNX Runtime API

C++

Architecture

ARM64

Execution Provider

Vitis AI

Execution Provider Library Version

No response

[yufenglee]

@jywu-msft, @BoarQing , could you please take a look?

[BoarQing]

I will discuss it with my colleague tomorrow

[farah-rostom98]

Waiting for your reply!

[github-actions]

This issue has been automatically marked as stale due to inactivity and will be closed in 7 days if no further activity occurs. If further support is needed, please provide an update and/or more details.