Annotation workflow

[murlock1000]

(WIP) Implementation of #216

Features:

• Demo Jupyter notebook showcasing functionality of the workflow
• Hot-reloading of splat and annotation data using a custom websocket protocol and proxy server.
• Annotations manager
• Selecting and toggling labelled splats
• Selected splat bounding box highlight
• Annotation visualization in centers mode with 'Show Classes' toggle in view options menu
• (WIP) Add/remove new labels
• (WIP) Add/remove selected splats to a certain label
• Class ID histogram
• Additional RGB histograms

Known Bugs:

• Histogram merges 0 and 1 Class IDs under single bucket
• Efficiency improvement: lack of dynamic shader initialization upon label upload
• Partial support for annotations - hardcoded dataset sample count = 1
• Class color shader not rerendering upon hot-reload

[slimbuck]

Wow, amazing @murlock1000. I guess this runs on windows? I'll definitely give a try when I can!

[murlock1000]

@slimbuck yes, it runs on windows with the environment prepared according to the gaussian-splatting project requirements - but that is only required to run the demo, not supersplat itself. Additionally, it only requires 'from scene.gaussian_model import GaussianModel', so the whole project dependency could be omitted by extracting the GaussianModel class from it directly and installing libraries like torch on the go.