Open hohonuuli opened 1 year ago
Conversion of localization shapes (circles, polygons) to be compatible with FathomNet (rectangles); enable storage of other localization types within FathomNet metadata
Related to #107
Per #105, build support for rich text descriptions for using Visual Language Models (a la Grounding DINO).
Important piece for 2024
Priority is segmentation masks. Will require backend and frontend work.
We could store segmentation masks using a grayscale PNG with the same dimensions as the source image. But if we need more than 255 masks per image (with 0 being no-mask), we would need to get creative.
e.g. to support CoralNet