FaBR Precise157 is a 157-image hard-case bird-mask benchmark for post-mask
correction and fine-structure semantic segmentation.
This dataset is the formal benchmark used in the FaBR paper experiments. It is
not the full 5,000+ image source annotation project.
Code for the FaBR mainline pipeline is available here:
ShallM/fabr-mainline
The benchmark is meant to be small, readable, and easy to inspect. It contains
the hard cases used for paper-facing evaluation… See the full description on the dataset page:
https://huggingface.co/datasets/MitchelMi/fabr-precise157.
