PDWN: Pyramid Deformable Warping Network for Video Interpolation

Code for our IEEE Open Journal of Signal Processing paper: "PDWN: Pyramid Deformable Warping Network for Video Interpolation (https://ieeexplore.ieee.org/document/9416770)"

Table of Contents

• demos
• requirements
• train
• test
• evaluate on videos

demos

Requirements

• Ubuntu
• Pytorch
• Cuda (10.1) & Cudnn (10.1)
• mmcv-full (https://github.com/open-mmlab/mmcv. please follow the guidence to install mmcv properly.)
• ffmpeg

Dataset

Vimeo-triplet can be downloaded from http://data.csail.mit.edu/tofu/dataset/vimeo_triplet.zip

Train

To train your own model, please use the following command:

python train.py --name experiment --dataroot [PATH TO THE DATASET] --dataset vimeo_tri  --model deform --kernel 3 --loss L1 --batch_size 32 --use_cuda True

Test

To replicate the results presented in the paper, please use the following command (Model is saved under ./checkpoints/vimeo_plus_single_no_norm_crop)

python test.py --name vimeo_plus_single_no_norm_crop --dataroot [PATH TO THE DATASET] --ensemble True --kernel 3 --model_load latest --result_path ./results --checkpoint_path ./checkpoints --dataset vimeo_tri

Evaluate on videos

python seq_eval.py --video_path ./sunflower_1080p25.mp4 --name vimeo_plus_single_no_norm_crop --model deform --kernel 3 --t_interp 2