Segmentation Fault while using Cuda Executioner Provider

[vaghawan]

Hi,

I'm trying to use insightface model inside pipeless. I've process.py where I write everything for the inference, while this works fine using CPU, when I switch to Cuda execution, I'm thrown with segmentation fault, in fact the stages do not load themselves. So we're halted on the very first step.

The onnx model loading part is added in init.py as
model = insightface.model_zoo.get_model('det_500m.onnx', providers=['CPUExecutionProvider']) . It is working fine with 'CPUExecutionProvider'.
But when the provider is changed to 'CUDAExecutionProvider', it is showing Segmentation fault.

Any help to figure out this would be much much appreciated. We've run the same script away from pipeless and it works just fine.

Thanks
V

[miguelaeh]

Hi @vaghawan ,
Try downloading the model and using a process.json to load it instead of using an init file

[vaghawan]

Which means, I can't have process.py right? Insightface has multiple models working for different tasks.

[miguelaeh]

you can concatenate as many stages as you want, using a model per stage