README.md

TensorRT MIRNet

Description

This sample contains code that convert TensorFlow Lite MIRNet model to ONNX model and performs TensorRT inference on TensorRT Container.

1. Pull TensorRT Container and run container.
2. Download TensorFlow Lite MIRNet Model from PINTO_model_zoo.
3. Convert to ONNX Model.
4. Convert ONNX Model to Serialize engine and inference.

Reference

• soumik12345/MIRNet
• sayakpaul/MIRNet-TFLite-TRT
• PINTO0309/PINTO_model_zoo

Environment

• PC
  • NVIDIA GPU
  • Linux
  • Docker or Podman
  • NVIDIA Container Toolkit

Pull and run NVIDIA TensorRT container

Run the NVIDIA Tensor RT container in Docker or Podman.
Requires TensorRT 8 or above (8/26 21.07-py3).

sudo podman run -it --rm nvcr.io/nvidia/tensorrt:21.07-py3

Install OpenCV-Python

apt update
apt install python3-opencv

Convert ONNX Model

Download TensorFlow Lite model

Clone PINTO_model_zoo repository and download MIRNet model.

git clone https://github.com/PINTO0309/PINTO_model_zoo.git
cd PINTO_model_zoo/079_MIRNet/
./download_INPUT_SIZE.sh

Convert ONNX Model

Install onnxruntime and tf2onnx.

pip3 install onnxruntime tf2onnx tensorflow

Convert TensorFlow Lite Model to ONNX Model.

python3 -m tf2onnx.convert --opset 13 \
    --tflite ./mirnet_INPUT_SIZE_float32.tflite \
    --output ./mirnet_INPUT_SIZE_float32.onnx

Check trtexec

trtexec --onnx=./mirnet_INPUT_SIZE_float32.onnx

Run demo

Clone this repository.

cd /workspace/
git clone https://github.com/NobuoTsukamoto/tensorrt-examples

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 /workspace/PINTO_model_zoo/079_MIRNet/mirnet_INPUT_SIZE_float32.onnx \
    --output /workspace/tensorrt-examples/models/mirnet_INPUT_SIZE.trt \
    --fp16

Finally you can run the demo.

python3 trt_mirnet_image.py \
    --model /workspace/tensorrt-examples/models/mirnet_INPUT_SIZE.trt
    --image PATH_TO_INPUT_IMAGE_FILE
    --output PATH_TO_OUTPUT_IMAGE_FILE
    --input_shape INPUT_SIZE(ex. 416,416)