Awesome

ECFVI : Error Compensation Framework for Flow-Guided Video Inpainting

This repository is for ECFVI introduced in the following paper

Jaeyeon Kang, Seoung Wug Oh, and Seon Joo Kim. "Error Compensation Framework for Flow-Guided Video Inpainting ", ECCV 2022. PDF

Dependencies

Python>=3.6.8, Pytorch=1.6.1, CUDA version>= 10.1 

Quickstart (DEMO)

1. Clone this github repo

   git clone https://github.com/JaeYeonKang/ECFVI
   cd ECFVI
    
    
   

2. Place your test dataset on ./test

   frame_path = %DATA_DIR% + %DATA_NAME% + 'frames/'
        ex) ./test/Youtube-VI/frames/video1/00000.png'

   mask_path = %DATA_DIR% + %DATA_NAME% + 'masks/'

3. Download our pretrained models from link. Then, place the weights in ./pretrained_weights

4. Run demo

   CUDA_VISIBLE_DEVICES=0 python main.py --model network --trainer core.evaluation --version $SAVE_DIR$ \
   --test_data_name $DATA_NAME$ --test_data_root $DATA_DIR$ --mask_mode $MASK_MODE$ 
   
   

   • SAVE_DIR : path to save results
   • DATA_NAME : name of test dataset
   • DATA_DIR : path to test dataset
   • MASK_MODE : for Youtube-VI dataset, mode:0 -> moving mask, mode:1 -> stationary mask

   For example,

   CUDA_VISIBLE_DEVICES=0 python main.py --model network --trainer core.evaluation --version DEMO \
   --test_data_name Youtube-VI --test_data_root ./test --mask_mode 0
   
   CUDA_VISIBLE_DEVICES=0 python main.py --model network --trainer core.evaluation --version DEMO \
   --test_data_name DAVIS_shadow --test_data_root ./test --davis
   
   

Youtube Video Inpainting (Youtube-VI) dataset

You can download our Youtube-VI dataset from link

Citation

If you use any part of this code in your research, please cite our paper