TinyTensor is an efficient lightweight deep learning inference framework.
TinyTensor supports a variety of popular neural network architectures such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and fully connected neural networks, and it can be used for tasks such as image classification, object detection, speech recognition, and natural language processing.
- Development language: C++ 20
- Math Library: Armadillo
- Logging framework：Google glog
- Unit test: Google Test
- Code style: Clang format
- Performance testing： Benckmark
apt update
apt install cmake libopenblas-dev liblapack-dev \
libarpack2-dev libsuperlu-dev libomp-dev libopencv-dev
wget https://sourceforge.net/projects/arma/files/armadillo-12.2.0.tar.xz
mkdir build && cd build
cmake ..
make -j8
make install
cd third_party
git submodule update --init
mv googletest benchmark
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=RELEASE ../benchmark
make -j8
# 如果想全局安装就接着运行下面的命令
sudo make install
- ReLU
- Sigmoid
- Conv
- MaxPooling
Intel(R) Xeon(R) W-2223 CPU @ 3.60GHz
gcc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
| Input size | Model | Computing Device | Time |
|---|---|---|---|
| 224×224 batch = 8 | ResNet18 | CPU(armadillo) | 55ms / image |
| 224×224 batch =16 | ResNet18 | CPU(armadillo) | 28.5ms / image |
caffe