2018-03-05, 19:42
- Name: Anurag Sarkar
- forum/e-mail: [email protected]
- Summary: I had a basic idea of using deep learning and specifically generative aderserial nets to generate new frames if there are missing frames in a video file, or use similar concept to create live tile of thee video a little different from just playing a small clip of the video itself . I mean it's just an idea and it'll be great if i can have feedback about it. i was doing a research on this subject for a while and was looking for a platform to implement this.
- How will I achieve this: Using deep learning and GAN's we will try to predict missing frames in the movie but in a little different fashion, most research that have been done are around predicting frames based on earlier frames, but i want to implement in using both future and present frames.
- What will the project focus on: i think it'll be an intuitive way to see how computers perform in generating new video files and maybe users can just put in random video to generate new video file(in a fun way of course)
- Benefits: integrating this to KODI will let user use movies with missing frames.
- Goals: I think this particular section is attainable in a couple of months i dont know how long will it take to integrate in the KODI software. the major sub goals can be getting the network working and then getting the accuracy right and then it can be implementing high res video files.
- What does it touch in Kodi: it can be a new space in the application where user may store their video files or images or anything they like and using our algorithm we can generate new video files or maybe make a story telling session for the user.
- Requirements: keras, Tensorflow, opencv, PIL and cuda
I've been coding for a good time now. I mostly work on machine learning and deep learning algorithms and i was looking to built something cool for the summers.
cheers,
Anurag
