Maskflownet.pytorch(updating...)

This is an unoffical implementation of MaskFlownet: Asymmetric Feature Matching with Learnable Occlusion Mask by using Pytorch.

Introduction

Requirements

• Pytorch 1.5.0(only for using DeformConv)
• CUDA 10.1+
• Tensorboard
• tqdm

Install

correlation

The correlation package must be installed first:

cd model/correlation_package
python setup.py install

DCN

if your pytorch is lower than 1.5.0, please install dcn package:

cd model/dcn
python setup.py install

Traing

Currenly, this repo only support the training pipeline of KITTI 2012/2015 and my own dataset.

sh tools/train.sh

Please modify the number of allocated gpus and the path of your configs before launch the whole project.

Inference