Topic: Image Quality Metrics Comparison

Hi,

I used to use SVP4 a lot, but have not used any video interpolation software that much recently. I have tried out the new RIFE model on some clips and found that it performed extremely well, and on Google Colab, it can easily do 2x 1080p interpolation at 9-10fps (which is insane for a neural network-based approach!). The authors of RIFE have published PSNR and SSIM results in their paper at https://arxiv.org/abs/2011.06294, and I was wondering if anyone has any data/results about how SVP (at the best default settings) compares to these neural net models. I might run SVP on these datasets when I get time if no one else does, but I figured I'd save time in case someone has already done this!

  • MAG79
  • SVP developer
  • Offline

Re: Image Quality Metrics Comparison

These metrics (PSNR and SSIM) must be calculated at the same source to compare them correctly.
We have not calculate PSNR and SSIM with videos by your link. It will be interesting if you will do it.