search

RangiLyu / nanodet

NanoDet-Plusāš”Super fast and lightweight anchor-free object detection model. šŸ”„Only 980 KB(int8) / 1.8MB (fp16) and run 97FPS on cellphonešŸ”„
Apache License 2.0
5.78k stars 1.04k forks source link

Cannot convert the onnx model into the tensorrt model with dynamic shape #453

Open frandoLin opened 2 years ago

frandoLin commented 2 years ago

image

Does anyone come across the same problem? For the batchsize of 1, the coversion is ok. Does that mean nanodet does not support dynamic inference by using tensorrt?

frandoLin commented 2 years ago

Whatever I try, the batchsize for the onnx model is fixed which is one.

image

menggui1993 commented 2 years ago

You need to export pytorch model to onnx model with dynamic_axes parameters