Ultra96v2，dpu have died

Support
1

Hello，I use Ultra96v2 and vitis ai to deploy my model. I can quantify and compile my model，and dpu can run normally.
But once the model input image local brightness is high, when I run “n2cube.dpuRunTask(task)”, the Jupyter will remind me：The kernel appears to have died. It will restart automatically.
I suspect that when the model is quantized, the quantization factor is too large, which leads to the overflow when the brightness is too high.
But how can I fix it?
My pynq-dpu is v2.5, vitis ai is v1.1.

2

If you look in /var/log/jupyter.log there may be more details on why the kernel died.

Peter

3

Hello, this is the error:
[DNNDK] Invalid pixel value of input tensor: 5782

I’ve done this: https://forums.xilinx.com/t5/AI-and-Vitis-AI/DNNDK-Invalid-pixel-value-of-input-tensor/td-p/1067862
but it still prompts for errors.
If I don’t multiply the input scale, the error is:
[DNNDK] Invalid pixel value of input tensor: 128

So, what should I do?

4

This is my problem: https://forums.xilinx.com/t5/AI-and-Vitis-AI/DNNDK-Invalid-pixel-value-of-input-tensor/td-p/1067862