Come up with method for reducing manual effort in image matting

[brysonjones]

• Current SotA methods for Image Matting require a "Trimap" of an image before extracting the foreground object
  • This requires manual creation of the trimap, which kinda defeats the purpose -- I get that it's a challenging problem, but manually outlining the foreground object is painful and reduces value proposition
• Current path forward is to just use images that already have the trimap
• Next step is building into the UI a way to easily generate a trimap for an image
• Final step is remove the need for trimap -- could use NeRF model output or something similar as an proxy

[brysonjones]

See: https://arxiv.org/pdf/2305.15272.pdf -- this is SoTA right now