This project is an official MegEngine implementation of A Dynamic Multi-Scale Voxel Flow Network for Video Prediction.
Our paper is accepted by CVPR2023.
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/*
Downloads Cityscapes and KITTI.
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