Skip to content

MegEngine/MegEngine-DMVFN

Repository files navigation

A Dynamic Multi-Scale Voxel Flow Network for Video Prediction (MegEngine implementation)

Introduction

This project is an official MegEngine implementation of A Dynamic Multi-Scale Voxel Flow Network for Video Prediction.

Our paper is accepted by CVPR2023.

Usage

Installation

git clone https://github.com/MegEngine/MegEngine-DMVFN.git
cd MegEngine-DMVFN
pip3 install -r requirements.txt
  • Download the pretrained models from Google Drive.

  • Unzip and move the pretrained parameters to pretrained_models/*

Installation

Downloads Cityscapes and KITTI.

Run

Train

python3 train_city_vgg.py --train_dataset CityTrainDataset --val_datasets CityValDataset --batch_size 8 --num_gpu 8

Test

Download CityValidation

python3  test.py --val_datasets CityValDataset --load_path ./pretrained_models/dmvfn_133.pkl

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages