2019/12/13 Report
===
## What's Done
- Our AI deinterlacing algorithm outperforms classical algorithms including Line Doubling and YADIF in terms of human vision assessments and numerical evaluations(PSNR and SSIM).
- The algorithm takes less than 5 milliseconds to produce 2 progressive frames by referencing 1 interlaced frame (two inter-weaved fields) and the time cost is still being improved.
## In Progress
- The above mentioned development is built on Linux and Python environment. We are migrating it into Windows and C++ and creating a DLL and test applications.
## What's Next
- We will integrate the AI deinterlacing functionality into Back End by supplying the DLL and C++ interfaces.
## For more information, please refer to
- [PoC #4 - Generate Datasets for Model Training and Train Deinterlacing Models](https://id4tvdev.atlassian.net/wiki/spaces/AI/pages/102989825/PoC+4+-+Generate+Datasets+for+Model+Training+and+Train+Deinterlacing+Models)
- [PoC #5 - Create a Dynamic-Link Library (DLL) for Using AI Deinterlacing Algorithms](https://id4tvdev.atlassian.net/wiki/spaces/AI/pages/115605505/PoC+5+-+Create+a+Dynamic-Link+Library+DLL+for+Using+AI+Deinterlacing+Algorithms)