README.md

TensorRT Fast-SCNN

Description

This sample contains code that performs TensorRT inference on Jetson.

1. Download ONNX Fast-SCNN Model from PINTO_model_zoo.
2. Convert ONNX Model to Serialize engine and inference on Jetson.

Reference

• PINTO0309/PINTO_model_zoo
  • 228_Fast-SCNN
  • Demo projects

Howto

Download ONNX Model

Clone PINTO_model_zoo repository and download Fast-SCNN model.

git clone https://github.com/PINTO0309/PINTO_model_zoo.git
cd PINTO_model_zoo/228_Fast-SCNN/
./download.sh

Check trtexec

/usr/src/tensorrt/bin/trtexec --onnx=./fast_scnn_NNNxNNN/fast_scnn_NNNxNNN.onnx

Convert ONNX Model to TensorRT Serialize engine file.

Copy fast_scnn_NNNxNNN.onnx to tensorrt-examples/models.
In the following, fast_scnn_576x768.onnx is taken as an example.

cp ~/home/nobuo/Data/models/PINTO_model_zoo/228_Fast-SCNN/fast_scnn_576x768/fast_scnn_576x768.onnx ~/tensorrt-examples/models/

Convert to Serialize engine file. If you want to convert to FP16 model, add --fp16 to the argument of convert_onnxgs2trt.py.

cd ~/tensorrt-examples/python/utils
python3 convert_onnxgs2trt.py \
    --model /home/jetson/tensorrt-examples/models/fast_scnn_576x768.onnx \
    --output /home/jetson/tensorrt-examples/models/fast_scnn_576x768.trt \

Finally you can run the demo.

python3 trt_fast_scnn_capture.py \
    --model ../../models/fast_scnn_576x768.trt
    --input_shape 576,768

or 

python3 trt_fast_scnn_image.py \
    --model ../../models/fast_scnn_576x768.trt
    --input_shape 576,768
    --input input_image.png
    --output output_image.png