start-cuda's Introduction

Start CUDA

Toolkit:

GPUs:

Books:

Build

sudo apt install -y build-essential cmake git

# depends (only for samples/img_proc)
# sudo apt install -y libboost-all-dev libjpeg-dev

# install OpenCV (required)
#  https://github.com/ikuokuo/start-opencv
export OpenCV_DIR=$HOME/opencv-4/lib/cmake

# install TensorRT (optional)
#  https://github.com/NVIDIA/TensorRT
export TensorRT_ROOT=/usr/local/TensorRT

git clone https://github.com/ikuokuo/start-cuda.git
cd start-cuda/
make

Tutorials

~~books~~
samples
- quick_start - CUDA 快速入门
- matmul
- npp
- nvjpeg
- tensorrt

start-cuda's People

Contributors

Stargazers

Watchers

start-cuda's Issues

A question.

In this fantastic write up: https://www.cnblogs.com/gocodinginmyway/p/15795663.html I was able to learn how to properly freeze a downsample hyperparameter. The resulting inference works brilliant.

Unfortunately I have been unable to replicate the inference logic found in: https://github.com/ikuokuo/start-cuda/blob/master/samples/tensorrt/rvm_infer.cc using the python TensorRT library. How might someone go about shaping the input/outputs to achieve this?

内存泄漏

作者您好，我参考了您的RVM的Tensorrt代码，但是我的项目中跑的是实时流，对于RVM的TensorRT代码中opencv那一块是存在内存泄漏问题的，我尝试释放每次创建的那些Mat矩阵，但是没什么效果，我是c++刚入门的，我想请教下您，关于opencv那块的代码您认为内存泄漏可能出现在哪里，感谢百忙之中查看，谢谢！

Recommend Projects