NVIDIA-AI-IOT
bevfusion cuda12 libspconv support？
I use orin jeston6.0 where cuda=12.2. can you provide libspconv aarch64 cuda12？ thank you
Me too, I use orin jeston6.0 where cuda=12.2., could you please provide libspconv aarch64 cuda12? should I create it by myself? and how?
Me too, I use orin jeston6.0 where cuda=12.2., could you please provide libspconv aarch64 cuda12? should I create it by myself? and how?
hello,I found this project tag deepsteam support cuda12 arm libspconv.so,but its speed is very slow.I test in cuda11 about 55ms ,but this libspconv.so about 110ms. also you can refer https://github.com/riteshkhrn/Lidar_AI_Solution/tree/main/libraries/New3DSparseConvolution to build your jeston6.0 lispocnv.so, BUT I failed .If you build success ,please share your libspconv.so cuda12 arm,this is very useful.
Me too, I use orin jeston6.0 where cuda=12.2., could you please provide libspconv aarch64 cuda12? should I create it by myself? and how?
hello,I found this project tag deepsteam support cuda12 arm libspconv.so,but its speed is very slow.I test in cuda11 about 55ms ,but this libspconv.so about 110ms. also you can refer https://github.com/riteshkhrn/Lidar_AI_Solution/tree/main/libraries/New3DSparseConvolution to build your jeston6.0 lispocnv.so, BUT I failed .If you build success ,please share your libspconv.so cuda12 arm,this is very useful.
I'm trying to create libspconv.so for cuda12 arm, It may needs some time. I will provide it's ready. Could you please tell me how to use the one in Deepstream? I copy it from the Deepstram branch bu it didn't work. Does it need to copy anything else with it.
Me too, I use orin jeston6.0 where cuda=12.2., could you please provide libspconv aarch64 cuda12? should I create it by myself? and how?
hello,I found this project tag deepsteam support cuda12 arm libspconv.so,but its speed is very slow.I test in cuda11 about 55ms ,but this libspconv.so about 110ms. also you can refer https://github.com/riteshkhrn/Lidar_AI_Solution/tree/main/libraries/New3DSparseConvolution to build your jeston6.0 lispocnv.so, BUT I failed .If you build success ,please share your libspconv.so cuda12 arm,this is very useful.
I'm trying to create libspconv.so for cuda12 arm, It may needs some time. I will provide it's ready. Could you please tell me how to use the one in Deepstream? I copy it from the Deepstram branch bu it didn't work. Does it need to copy anything else with it.
I think you should copy lidar-scn-onnx-parser.cpp lidar-scn-onnx-parser.hpp and onnx DIR which container .proto files and you should modify forward code as this.This is a centerpoint demo I modified in centerpoint.cpp.Hope you sucess!
#include <spconv/engine.hpp>
#include "lidar-scn-onnx-parser.hpp"
// auto result = scn_engine_->forward(
// {valid_num, 5}, spconv::DType::Float16, d_voxel_features,
// {valid_num, 4}, spconv::DType::Int32, d_voxel_indices,
// 1, sparse_shape, stream
// );
scn_engine_->input(0)->set_data(
{valid_num, 5}, spconv::DataType::Float16, d_voxel_features,
{valid_num, 4}, spconv::DataType::Int32, d_voxel_indices, sparse_shape
);
scn_engine_->forward(stream);
trt_->forward({scn_engine_->output(0)->features().ptr(), d_reg_[0], d_height_[0], d_dim_[0], d_rot_[0], d_vel_[0], d_hm_[0],
d_reg_[1], d_height_[1], d_dim_[1], d_rot_[1], d_vel_[1], d_hm_[1],
d_reg_[2], d_height_[2], d_dim_[2], d_rot_[2], d_vel_[2], d_hm_[2],
d_reg_[3], d_height_[3], d_dim_[3], d_rot_[3], d_vel_[3], d_hm_[3],
d_reg_[4], d_height_[4], d_dim_[4], d_rot_[4], d_vel_[4], d_hm_[4],
d_reg_[5], d_height_[5], d_dim_[5], d_rot_[5], d_vel_[5], d_hm_[5]}, stream);
Me too, I use orin jeston6.0 where cuda=12.2., could you please provide libspconv aarch64 cuda12? should I create it by myself? and how?
hello,I found this project tag deepsteam support cuda12 arm libspconv.so,but its speed is very slow.I test in cuda11 about 55ms ,but this libspconv.so about 110ms. also you can refer https://github.com/riteshkhrn/Lidar_AI_Solution/tree/main/libraries/New3DSparseConvolution to build your jeston6.0 lispocnv.so, BUT I failed .If you build success ,please share your libspconv.so cuda12 arm,this is very useful.
I'm trying to create libspconv.so for cuda12 arm, It may needs some time. I will provide it's ready. Could you please tell me how to use the one in Deepstream? I copy it from the Deepstram branch bu it didn't work. Does it need to copy anything else with it.
hello，have you successfully built libspconv for aarch64 with CUDA 12? Could you share it with me, please?
Me too, I use orin jeston6.0 where cuda=12.2., could you please provide libspconv aarch64 cuda12? should I create it by myself? and how?
hello,I found this project tag deepsteam support cuda12 arm libspconv.so,but its speed is very slow.I test in cuda11 about 55ms ,but this libspconv.so about 110ms. also you can refer https://github.com/riteshkhrn/Lidar_AI_Solution/tree/main/libraries/New3DSparseConvolution to build your jeston6.0 lispocnv.so, BUT I failed .If you build success ,please share your libspconv.so cuda12 arm,this is very useful.
I'm trying to create libspconv.so for cuda12 arm, It may needs some time. I will provide it's ready. Could you please tell me how to use the one in Deepstream? I copy it from the Deepstram branch bu it didn't work. Does it need to copy anything else with it.
hello，have you successfully built libspconv for aarch64 with CUDA 12? Could you share it with me, please?
refer to this repo: https://github.com/xiongda777/bevfusion_ros.git
Me too, I use orin jeston6.0 where cuda=12.2., could you please provide libspconv aarch64 cuda12? should I create it by myself? and how?
hello,I found this project tag deepsteam support cuda12 arm libspconv.so,but its speed is very slow.I test in cuda11 about 55ms ,but this libspconv.so about 110ms. also you can refer https://github.com/riteshkhrn/Lidar_AI_Solution/tree/main/libraries/New3DSparseConvolution to build your jeston6.0 lispocnv.so, BUT I failed .If you build success ,please share your libspconv.so cuda12 arm,this is very useful.
I'm trying to create libspconv.so for cuda12 arm, It may needs some time. I will provide it's ready. Could you please tell me how to use the one in Deepstream? I copy it from the Deepstram branch bu it didn't work. Does it need to copy anything else with it.
hello，have you successfully built libspconv for aarch64 with CUDA 12? Could you share it with me, please?
refer to this repo: https://github.com/xiongda777/bevfusion_ros.git
hi, is your bevfusion spconv inference speed normal? I've tested the latest official code on jetson orin,it's very slow, taking 110ms.
Me too, I use orin jeston6.0 where cuda=12.2., could you please provide libspconv aarch64 cuda12? should I create it by myself? and how?
hello,I found this project tag deepsteam support cuda12 arm libspconv.so,but its speed is very slow.I test in cuda11 about 55ms ,but this libspconv.so about 110ms. also you can refer https://github.com/riteshkhrn/Lidar_AI_Solution/tree/main/libraries/New3DSparseConvolution to build your jeston6.0 lispocnv.so, BUT I failed .If you build success ,please share your libspconv.so cuda12 arm,this is very useful.
I'm trying to create libspconv.so for cuda12 arm, It may needs some time. I will provide it's ready. Could you please tell me how to use the one in Deepstream? I copy it from the Deepstram branch bu it didn't work. Does it need to copy anything else with it.
hello，have you successfully built libspconv for aarch64 with CUDA 12? Could you share it with me, please?
refer to this repo: https://github.com/xiongda777/bevfusion_ros.git
hi, is your bevfusion spconv inference speed normal? I've tested the latest official code on jetson orin,it's very slow, taking 110ms.
Testing on our machines shows no significant difference in speed.