my project have four 'HEAD_CLS_NAME',and how to modify the code to run success in program,expecially the postprocess.cu;

hova88 / PointPillars_MultiHead_40FPS

A REAL-TIME 3D detection network [Pointpillars] compiled by CUDA/TensorRT/C++.

GNU General Public License v3.0

276 stars 87 forks source link

my project have four 'HEAD_CLS_NAME',and how to modify the code to run success in program,expecially the postprocess.cu; #10

Closed qiuhui1991 closed 3 years ago

qiuhui1991 commented 3 years ago

RPN_HEAD_CFGS: [ { 'HEAD_CLS_NAME': [ 'vehicle' ], }, { 'HEAD_CLS_NAME': [ 'car' ], }, { 'HEAD_CLS_NAME': [ 'pedestrian'], }, { 'HEAD_CLS_NAME': [ 'cyclist' ], }, ]

hova88 commented 3 years ago

RPN_HEAD_CFGS: [ { 'HEAD_CLS_NAME': [ 'vehicle' ], }, { 'HEAD_CLS_NAME': [ 'car' ], }, { 'HEAD_CLS_NAME': [ 'pedestrian'], }, { 'HEAD_CLS_NAME': [ 'cyclist' ], }, ]

这个说起来有点复杂，而且我写的postprocess.cu还有很多冗余和不足的地方，你可以看看apollo或者autoware的postprocess部分，他们是singlehead的，对比这我这个，你可以设计一个符合自己需求的后处理单元。

qiuhui1991 commented 3 years ago

RPN_HEAD_CFGS: [ { 'HEAD_CLS_NAME': [ 'vehicle' ], }, { 'HEAD_CLS_NAME': [ 'car' ], }, { 'HEAD_CLS_NAME': [ 'pedestrian'], }, { 'HEAD_CLS_NAME': [ 'cyclist' ], }, ]

这个说起来有点复杂，而且我写的postprocess.cu还有很多冗余和不足的地方，你可以看看apollo或者autoware的postprocess部分，他们是singlehead的，对比这我这个，你可以设计一个符合自己需求的后处理单元。

谢谢