xjock / deepstream-openpose

deepstream-openpose

1. Run CMU pose demo

Platform: xavier, Jetpack 4.3 with DeepStream 4.0.2

Notes:

1. Porting to Tesla/x86 platform should be easy.
2. CUDA_cublas_device_LIBRARY NOTFOUND issue -> Solution
3. Refer to openpose/scripts/ubuntu/install_deps.sh to install deps libs.
4. Refer to openpose/models/getModels.sh to fetch models
5. Build.

   $ cmake -D CMAKE_BUILD_TYPE=Debug ..
   $ make -j4
   

6. These demos can work.

   $ ./build/examples/openpose/openpose.bin
   $ ./build/examples/tutorial_api_cpp/01_body_from_image_default.bin
   $ ...
   

It show like this.

2. Deploy pose coco model by DeepStream 4.0.2

Model: pose/coco/pose_iter_440000.caffemodel, pose/coco/pose_deploy_linevec.prototxt

Pipeline:

filesrc -> jpegparse -> nvv4l2decoder -> nvstreammux -> nvinfer (openpose and 18 parts parse) nvsegvidsual -> nvmultistreamtiler -> (nvegltransform) -> nveglglessink

Build libnvds_infer.so

$ cd libs/nvinfer
$ make
Backup /opt/nvidia/deepstream/deepstream-4.0/lib/libnvds_infer.so
$ sudo ln -sf $(pwd)/libnvds_infer.so /opt/nvidia/deepstream/deepstream-4.0/lib/libnvds_infer.so 

Build openpose-app

$ cd openpose_app
$ make
Change nvinfer_config.txt "model-file" and "proto-file" to be your path
 model-file=
 proto-file=

Run

$ ./openpose-app ./nvinfer_config.txt COCO_val2014_000000000564.jpg

3. TODO

Add dsexample plugin after nvinfer and do resize_merge, nms and BodyPartConnector, and show result by nvosd like the below.