Inspiration
Some of the thoughts were from the point of one the client's issue I've tackled while freelancing. The client explained me the problem of long rendering process in the company.
What it does
There are a few existed algorithms, which interpolate frames. The best good experience is with RIFE. RIFE gives us a chance to interpolate 2X frame on regular animations (video). It's necessary to increase the interpolation multiplier to 5-10X. For that, the conversion of 5FPS video to 50FPS video is needed. RIFE doesn't work for this task because of a lot of artefacts with so high interpolation multiplier.
How we built it
We decided that we can make RIFE faster providing the depth maps instead of RGB images to the algorithm and then colorising them with the use of Neural Networks
Challenges we ran into
Research on datasets and computational resources for training
Accomplishments that we're proud of
First results of colorising depth maps are nice
What we learned
Never give up and find alternative ways to solve a problem
What's next for Frame Interpolation App
1) Establishing a company in Krakow, Poland 2) Finding resources and large servers on a suitable costs for additional training 3) Providing the app to the current clients (2 of them), which are on hold and waiting for a solution 4) Designing a logo and a cool name for a product 5) Start running a landing website of the product 6) Increase sales
Log in or sign up for Devpost to join the conversation.