So I've been working on this project for quite some time now, my aim is to improve the video quality of old low-resolution low fps videos using computer vision and neural networks. For now the full list of stuff I use:
OpenCV and C++,https://github.com/tensorlayer/srganhttps://github.com/sniklaus/pytorch-s
Video with old and new cinematic at the same time (new on the top, old on the bottom): https://youtu.be/UjJlKUa5GRI
For lovers of the russian language here is the enhanced video with russian voiceover: https://youtu.be/05HojCkXX1I
(please view in 1080p60fps)
This is the first version and there is still a lot to improve (like the random jitters on the right and overall blurriness of the video). As the first step, I am going to move from SRN-Deblur to DeblurGAN implementation.
Hope you enjoy what you see :)
So this is the continuation of my project (V2) to enhance the cinematics of the old games. From the previous version, there is now a significant improvement in quality and stability.
You can compare the original cinematic and enhanced one here: https://youtu.be/R9jXwm_CTDM
I also uploaded the human ending cinematic enhanced using new approach (V2): https://youtu.be/VREdwGHJls8
And if you want to compare a V2 human ending cinematic to original one, it's here: https://youtu.be/TPYRxQ9Ct4o
For now my biggest issue is dealing with motion blur, any heavily moving scene does not get much of improvement due to the blur and all the solutions I use for now reduce it, but not to the desired degree, so there is still a lot of work to do.
If you have on mind other game's cinematics which you want to see enhanced - let me know, I'm looking forward to continue working on this project, but I want to have a bigger set of videos to work with.