A TensorRT implementation of EfficientNet. For the Pytorch implementation, you can refer to EfficientNet-PyTorch
- install
efficientnet_pytorch
pip install efficientnet_pytorch
- gennerate
.wts
file
python gen_wts.py
- build
mkdir build
cd build
cmake ..
make
- serialize model to engine
./efficientnet -s [.wts] [.engine] [b0 b1 b2 b3 ... b7] // serialize model to engine file
such as
./efficientnet -s ../efficientnet-b3.wts efficientnet-b3.engine b3
- deserialize and do infer
./efficientnet -d [.engine] [b0 b1 b2 b3 ... b7] // deserialize engine file and run inference
such as
./efficientnet -d efficientnet-b3.engine b3
- see if the output is same as pytorch side
For more models, please refer to tensorrtx