Skip to content

README.md

Latest commit

 

History

History
89 lines (71 loc) · 2.32 KB

README.md

File metadata and controls

89 lines (71 loc) · 2.32 KB

MeFlow

How to Run (Pytorch)

  1. set up a virtual environment.
cd MeFlow
git clone https://github.com/gangweix/MeFlow.git
cd MeFlow

# Create a new conda environment with Python 3.11
conda create -n meflow -y python=3.11

# Activate the created environment
conda activate meflow

# Install the required Python packages
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
pip install gdown
pip install tensorboard
pip install opencv-python
pip install scipy
pip install pillow==9.5.0
  1. download pretrained checkpoints.
# things.pth
gdown --fuzzy https://drive.google.com/file/d/1_Ug6FZhK1V0spDPu578QG65uxTJ-ENQL/view?usp=drive_link -O pretrained_models/

# sintel.pth
gdown --fuzzy https://drive.google.com/file/d/1AcHbcWHRja-0IzxMVxZv9YbpCmI4L_EW/view?usp=drive_link -O pretrained_models/

# kitti.pth
gdown --fuzzy https://drive.google.com/file/d/1KuwYWOnJakECsPKOl6nj2E4V7bpJVH0j/view?usp=drive_link -O pretrained_models/

# chairs.pth
gdown --fuzzy https://drive.google.com/file/d/1NxBqOmpuxtVAA_R5GzNYdRXLLFoK2-9m/view?usp=drive_link -O pretrained_models/
  1. run the original pytorch model on test images.
bash ./scripts/demo.sh
  1. check pytorch model inference performance
cd ..
python ../gen_video2imgs.py
python infer.py

  • input size: 288 x 512, iters: 20

  • original

    • 249 iterations time: 60.6272 [sec]
    • Average FPS: 4.11 [fps]
    • Average inference time: 243.48 [msec]

  • wrapper for onnx/trt

    • 249 iterations time: 192.1871 [sec]
    • Average FPS: 1.30 [fps]
    • Average inference time: 771.84 [msec]

How to Run (TensorRT)

  1. generate onnx file
python onnx_export.py
// a file '.onnx' will be generated in onnx directory.
  1. build tensorrt model and run
conda activate trte
python onnx2trt.py
// a file '.engine' will be generated in engine directory.
  • input size: 288 x 512, iters: 20
  • Engine build done! (197.0 [min] 7.06 [sec])
  • 249 iterations time: 13.4447 [sec]
  • Average FPS: 18.52 [fps]
  • Average inference time: 53.99 [msec]

Back