Pytorch : Run FlowNet2 with Pytorch

Method

1. Install Ubuntu16.04
2. Install NVIDIA Driver and CUDA 10.0
3. Install Anaconda python3.7
4. Change Anaconda Mirror
5. Install Pytorch 1.0.1 with CUDA 10.0

conda install pytorch==1.0.1 torchvision==0.2.2 cudatoolkit=10.0 -c pytorch

6. git clone nvidia: flownet2-pytorch
7. use conda to install all the necessary package
8. install flownet2-pytorch/networks use code

bash install.sh

update install.sh as flows:

#!/bin/bash
cd ./networks/correlation_package
rm -rf *_cuda.egg-info build dist __pycache__
python3 setup.py install 

cd ../resample2d_package
rm -rf *_cuda.egg-info build dist __pycache__
python3 setup.py install

cd ../channelnorm_package
rm -rf *_cuda.egg-info build dist __pycache__
python3 setup.py install

cd ..

9. replace from scipy.misc import imread, imresize to from imageio import imread
10. replace async=True to non_blocking=True in main.py
11. download model and MPISintel and predict

python main.py --inference --model FlowNet2 --save_flow --inference_dataset MpiSintelClean --inference_dataset_root '/home/yqs/data/flownet2-pytorch/dataset/MPI-Sintel/training'    --resume '/home/yqs/data/flownet2-pytorch/models/FlowNet2_checkpoint.pth.tar' 

12. show

import utils
import matplotlib.pyplot as plt
img = flow_utils.readFlow('/home/yqs/data/flownet2-pytorch/work/inference/run.epoch-0-flow-field/000000.flo')
plt.imshow(img[:,:,0])
plt.imshow(img[:,:,1])

---

CODE

https://github.com/Yannnnnnnnnnnn/flownet2-pytorch