Problems encountered when using the engine file to inference(mmrotate)

open-mmlab / mmdeploy

OpenMMLab Model Deployment Framework

https://mmdeploy.readthedocs.io/en/latest/

Apache License 2.0

2.71k stars 620 forks source link

Problems encountered when using the engine file to inference(mmrotate) #1313

Open wahaha opened 1 year ago

wahaha commented 1 year ago

📚 The doc issue

mmrotate's pth model has been converted to an engine file, but when I use the engine file for inference, only one class can be detected, and the rest are undetected

Suggest a potential alternative/fix

No response

wahaha commented 1 year ago

TensorRT

wahaha commented 1 year ago

When I used tensorRT to transform the model, the visualization of the tensorrt model yielded only one class

lvhan028 commented 1 year ago

Can you share the commands, the model config, and the checkpoint so that we can reproduce it?

wahaha commented 1 year ago

All document versions are as follows: (1) mmdeploy 0.10.0; (2) mmrotate 0.3.3; (3) mmcv-full 1.6.0; (4) ppl.cv 0.7.0; (5) TensorRT-8.2.3.0.Windows10.x86_64cuda-11.4.cudnn8.2; (6) CUDA11.1; (7) cudnn-11.3-windows-x64-v8.2.1.32; (8) pytorch 1.8.0; Whether there are problems with the above versions？