META has released an open-source AI model that segments and highlights objects in images. As the name 'Segment Anything' suggests, it handles any objects well.
Features:
1) Simple architecture + support for multimodality
2) Not only the model itself is open-sourced, but also the data used for training (11 million images + 1 billion defined segments)
Paper: https://scontent-lga3-1.xx.fbcdn.net/v/t39.2365-6/10000000_6331779526880473_6748528980292947838_n.pdf
Website: https://segment-anything.com/
Github: https://github.com/facebookresearch/segment-anything
P.S. It's incredibly pleasing to see that open-source principles continue to work.
P.S.S. For those unaware, open source is a format where your work (code, data, scientific discoveries) is published openly for free use by others. There are options where scientific and personal use is free, but commercial use requires a license purchase, though this is usually not considered true open source.