PlumageParts annotations and trained models for fine-grained avian plumage segmentation

This record contains the PlumageParts annotation release and trained model checkpoints associated with the manuscript “A fine-grained avian plumage segmentation dataset and benchmark for ecological image analysis”. The PlumageParts annotations correspond to 4,705 bird photographs selected from the iRateBirds Citizen Science Project, whose source photographs originate from the Macaulay Library. Original source images are not redistributed in this release. Instead, this record provides pixel-wise segmentation masks for nine ornithologically defined plumage regions: head, back, tail, throat, breast, belly, vent, coverts and remiges, plus background, together with metadata and train/validation/test split information. Masks are stored as single-channel PNG files and use filename stems based on scientific name, sex metadata and Macaulay photo catalogue ID. This release also includes trained model checkpoints, prediction masks and files used for external benchmark evaluation. External datasets such as CUB-200-2011 and PartImageNet should be obtained from their original sources and used under their respective licences. Full documentation, code and usage instructions are available at:https://github.com/EchanHe/PlumageParts