crosstyan/RapidPoseTriangulation
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Using tensorrt runtime directly.

Browse Source
This commit is contained in:
Daniel
2024-12-18 12:10:45 +01:00
parent 6e8f6a22ba
commit b26ec998b3
2 changed files with 145 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
extras/mmdeploy/README.md
View File

@ -49,6 +49,28 @@ python3 /RapidPoseTriangulation/extras/mmdeploy/add_extra_steps.py
<br>
## TensorRT
Run this directly in the inference container (the TensorRT versions need to be the same)
```bash
export withFP16="_fp16"
trtexec --fp16 \
--onnx=/RapidPoseTriangulation/extras/mmdeploy/exports/rtmdet-nano_320x320"$withFP16"_extra-steps.onnx \
--saveEngine=end2end.engine
mv ./end2end.engine /RapidPoseTriangulation/extras/mmdeploy/exports/rtmdet-nano_1x320x320x3"$withFP16"_extra-steps.engine
trtexec --fp16 \
--onnx=/RapidPoseTriangulation/extras/mmdeploy/exports/rtmpose-m_384x288"$withFP16"_extra-steps.onnx \
--saveEngine=end2end.engine
mv ./end2end.engine /RapidPoseTriangulation/extras/mmdeploy/exports/rtmpose-m_1x384x288x3"$withFP16"_extra-steps.engine
```
<br>
## Benchmark
```bash