Video Frame Interpolation Benchmark

This is the repository for evaluating video frame interpolation (VFI) methods. These codes are used to produce metrics in "PerVFI".

In this repository, we include evaluation of both image datasets and video datasets. For image datasets, we support PSNR, SSIM, LPIPS, DISTS, and IE metrics. For video datasets, we support not only those metrics for image datasets, but also FloLPIPS and VFIPS.

Usage

Build models and metrics

1. Please download the model checkpoints into the checkpoints folder, and change the filepath in the build_models.py file.

2. Download the datasets to the data/NAME-OF-DATASET folder, and modify the corresponding dataloader in tools.py.

3. Run the scripts:

   Evaluate image datasets:

   python evaluate_imageDst.py \
       -m MODEL-NAME \
       -dst DATASET-NAME
   

   Evaluate video datasets:

   python evaluate_imageDst.py \
               -m MODEL-NAME \
               -dst DATASET-NAME
   

To locate supported MODEL-NAME, please refer to build_models.py.

To locate supported DATASET-NAME, please refer to evaluate_imageDst.py and evaluate_videoDst.py.

Future:

This repository is under refinement. Please feel free to contribute or suggest improvements.