MMDetection2系列-(3)-部署

1. Pytorch to ONNX

pip install onnx-simplifier 
pip install onnxruntime
python tools/deployment/pytorch2onnx.py configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py work_dirs/faster_rcnn_r50_fpn_1x_coco/epoch_10.pth --output-file work_dirs/faster_rcnn_r50_fpn_1x_coco/epoch_10.onnx --show --verify --simplify

2. ONNX to TensorRT

• 安装tensorrt
  参考:deepstream 学习笔记

python tools/deployment/onnx2tensorrt.py work_dirs/mask_rcnn_r50_fpn_1x_coco/mask_rcnn_r50_fpn_1x_coco.onnx --trt-file work_dirs/mask_rcnn_r50_fpn_1x_coco/mask_rcnn_r50_fpn_1x_coco.trt --input-img demo/demo.jpg --shape 400 600 --mean 123.675 116.28 103.53 --std 58.395 57.12 57.375 --show --verify

---

参考:https://mmdetection.readthedocs.io/