linghu8812 / tensorrt_inference

708 stars 208 forks source link

How to export mxnet Retinaface model with dynamic input size? #166

Closed cyrusbehr closed 1 year ago

cyrusbehr commented 1 year ago

I have followed your instructions here and was able to convert the mxnet Retinaface model to onnx format. When the model is exported to onnx, it has a fixed input size of 640x640:

image

I know this model can work with dynamic input sizes, so how can it be exported such that the input size isn't hardcoded and it supports dynamic?