Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> rotoscoping is very, very time consuming

I feel like there is a cool machine learning project in here.



This is called image segmentation - Facebook recently open sourced DeepMask/SharpMask which does this https://github.com/facebookresearch/deepmask


I wonder if any of this has made it into vfx production? The few companies I've worked at they've invested on better tools to manually rotoscope because automated tools never lived up to the hype in production.

The most important thing in these masks are clean edges (and the "auto stuff" usually isn't). A rough mask (called a garbage matte) is good enough in some cases, but most often it's the first step and a clean matte is created by hand using multiple techniques; playing with contrast, playing with color/chroma (bluescreen), and hand animating curves or painting (rotoscoping).


I can't read the text, but this is the only video I can find that might be deepmask / sharpmask in action: https://www.youtube.com/watch?v=uOBND1_RNMg

If that is indeed a representation of the quality of results, it's clearly got a very long way to go before it's actually useful for rotoscoping. I suppose it could be useful for automating garbage mattes, but those are pretty easy to do anyway.




Consider applying for YC's Summer 2026 batch! Applications are open till May 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: