This repository implement the real-time Instance Segmentation Algorithm named SparseInst with TensoRT and ONNX.

• The initial repository on which I build mine is from hustvl/SparseInst__ repository (https://github.com/hustvl/SparseInst.git), for additional information about the installation of SparseInst, refer to the original repository.
• This project is built upon the excellent framework detectron2, and you should install detectron2 first, please check official installation guide for more details. (https://github.com/facebookresearch/detectron2.git)
• For command other than TensoRT and ONNX inference, please refer to the initial repository (e.g test_net.py).
• If you face any problem during the parsing time, don't hesitate to drop an issue or a ⭐ if there aren't any. if you have compatibility problem, check the model weights uploaded in the table below and go directly in the testing section.
• Be aware that in order to parse the model to ONNX and TensorRT, the files sparseinst.py, encoder.py and decoder.py has been modified/slightly modified, don't forget to check the modifications if you come from the initial repository.

pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116

If other versions of torch are needed, select yours by putting torch==1.11.0+cu102 for example.

sudo wget https://developer.download.nvidia.com/compute/cuda/repos/wsl-ubuntu/x86_64/cuda-wsl-ubuntu.pin
sudo mv cuda-wsl-ubuntu.pin /etc/apt/preferences.d/cuda-repository-pin-600
sudo wget https://developer.download.nvidia.com/compute/cuda/11.7.1/local_insta
    llers/cuda-repo-wsl-ubuntu-11-7-local_11.7.1-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-11-7-local_11.7.1-1_amd64.deb
sudo cp /var/cuda-repo-wsl-ubuntu-11-7-local/cuda-96193861-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda

python3 -m pip install --upgrade setuptools pip
python3 -m pip install nvidia-pyindex
python3 -m pip install --upgrade nvidia-tensorrt

Verify installation by writing  : assert tensorrt.Builder(tensorrt.Logger())

pip install onnxruntime-gpu
pip install onnxruntime
pip install numpy protobuf==4.21.5  
pip install onnx

The inference speed for Pytorch, ONNX and TensorRT has been compared and shown in the table below. SparseInst running with TensoRT achieved more a less 3 times faster inference speed of SparseInst than running with Pytorch. Lowering the input size of the image can lead to a decent real-time speed. The models from TensorRT and ONNX are built upon the first Pytorch listed weights in the table below : SparseInst R-50 G-IAM.

Note: All the computations has been done on a Nvidia Jetson TX2 Jetpack 4.6. Further test will be done on a Nvidia 2070 RTI

To build the model from Pytorch to ONNX, you need to run the following command. You can set the arguments to default. Please check if the config path and the model weights path are correctly set up.

<sudo python3 convert_onnx.py --config-file config-gile --output output_directory_onnxmodel --image dummy_input --opts MODEL.WEIGHTS weights_directory>

To build the model from ONNX to TensorRT, you need to run the following command. You can set the arguments to default. If you have any problem while parsing the model to TensorRT, don't hesitate to ask.

<sudo python3 convert_tensortt.py --onnx_model onnx-model-directory --output output_directory_TensoRTModel

To test the inference speed (FPS) of the Pytorch, TensorRT and ONNX models, run the following command.

1. Pytorch

sudo python3 eval_tensorrt_onnx.py  -c 0.2 --width_resized 320 --height_resized 320 --input datasets/coco/calib_images/*  --use_pytorch 

2. TensorRT

sudo python3 eval_tensorrt_onnx.py  -c 0.2 --width_resized 320 --height_resized 320 --input datasets/coco/calib_images/*  --use_tensorrt --tensorrt_engine engine/sparseinst_trt_320_320.engine

3. ONNX

sudo python3 eval_tensorrt_onnx.py  -c 0.2 --width_resized 320 --height_resized 320 --input datasets/coco/calib_images/* --use_onnx --onnx_engine onnx/sparseinst_onnx_320_320.onnx 

Notes :

• Input argument can either be an image or a directory of images (directory/*)

• You can of course infer all three together, just add the argument --use_model of the model you want to infer aswell as the engine (Not for Pytorch).

• In the terminal :

  • TRT inference only time reprensents the inference speed of the model alone

  TRT inference only use time 4.970773220062256 for 100 images, FPS=20.117594501474272
  

  • TRT algorithm time represents the inference speed and the preprocessing time combined

  TRT algorithm use time 22.519110441207886 for 100 images, FPS=4.440672745980644
  

To visualize segmentation results on your images, you can run the following commands :

1. Pytorch

sudo python3 eval_tensorrt_onnx.py  -c 0.2 --width_resized 320 --height_resized 320 --input datasets/coco/calib_images/*  --use_pytorch --output_pytorch results/result_image_pytorch/result_pytorch --save_image

2. TensorRT

sudo python3 eval_tensorrt_onnx.py  -c 0.2 --width_resized 320 --height_resized 320 --input datasets/coco/calib_images/*  --use_tensorrt --tensorrt_engine engine/sparseinst_trt_320_320.engine --output_tensorrt results/result_image_tensorrt/result_tensorrt --save_image

3. ONNX

sudo python3 eval_tensorrt_onnx.py  -c 0.2 --width_resized 320 --height_resized 320 --input datasets/coco/calib_images/* --use_onnx --onnx_engine onnx/sparseinst_onnx_320_320.onnx --output_onnx results/result_image_onnx/result_onnx --save_image

Notes :

• You can still infer and visualize all three together, just add all the argument together
• If you don't specify --save_image, it will only infer the model and not save the outputs.