Inference with NCNN and a custom model

I’ve been trying to get my VisionFive2 to do an inference on an image I get from a ROS2 Humble topic. I’ve tried numerous inferece firmwares, with no succes. The closest I’ve gotten is to make an inference with the example models from NCNN.
I’ve been trying to modify the original classes from these examples and stil didn’t get anything. Has anyone made an inference with their own models and NCNN?

1 Like

JH7110 is not include any NPU modules.

So, I think you need check the software is running right.
There have 2 way to run AI software.
1 is use float point instruction. But some AI-lib is not supported use common FP functions.
2 is use the IMG’s GPU.


The inference sotftware actually works, it’s just that it doesn’t work with the models I’ve trained. It makes an inference with the script i get from ultralytics when I convert my model to NCNN. But I can’t make it do an inference on an image.