README.md

Semantic Segmentation (BEN2:Background Erase Network)

How to Run

1. set up a virtual environment.

   git clone https://github.com/PramaLLC/BEN2.git
   cd BEN2
   conda create -n ben2 -y python=3.11
   conda activate ben2
   pip install git+https://github.com/PramaLLC/BEN2.git
   pip install onnxsim
   pip install onnx
   pip install onnxscript
   pip install opencv-python
   

2. download checkpoint

   mkdir -p checkpoint
   wget https://huggingface.co/PramaLLC/BEN2/resolve/main/BEN2_Base.pth -P checkpoint
   

3. check pytorch model inference

   cd ..
   python infer.py
   

How to Run (TensorRT)

1. generate onnx file

   python onnx_export.py
   

2. generate tensorrt model

   python onnx2trt.py
   

• input size : 512

  • [TRT_E] 100 iterations time: 3.6876 [sec]
  • [TRT_E] Average FPS: 27.12 [fps]
  • [TRT_E] Average inference time: 36.88 [msec]

• input size : 1024

  • [TRT_E] 100 iterations time: 11.9791 [sec]
  • [TRT_E] Average FPS: 8.35 [fps]
  • [TRT_E] Average inference time: 119.79 [msec]

• BEN: Using Confidence-Guided Matting for Dichotomous Image Segmentation

• BEN2 official GitHub