Datasets
Open-source benchmarking software and the restricted-access Amatis research dataset are presented as separate resources with separate rights.
Segmentation Benchmarking open-source software
A PyTorch research codebase for comparing two-stage semantic segmentation approaches under a shared protocol.
Amatis restricted-access wound dataset
A v0.1.0 wound-segmentation research package with 465 source records, 586 derived crop pairs and seven RGB mask classes, available only after reviewed approval.
Amatis Wound Segmentation Dataset v0.1.0
A restricted-access research dataset for pixel-level wound-image segmentation. It is separate from the public Segmentation Benchmarking software repository and is not available as a public download.
- source image–mask–annotation records
- 465
- derived wound-crop image–mask pairs
- 586
- RGB mask labels
- 7
- current dataset version
- v0.1.0
- What it contains
- The source set contains 448 PNG and 17 JPEG wound photographs, each paired with an authoritative RGB segmentation mask and a per-image JSON record. The package also includes derived wound-crop image–mask pairs.
- Research purpose
- The dataset supports approved non-commercial segmentation training and evaluation, benchmarking, preprocessing and reproducibility studies. It is not validated for diagnosis, treatment, triage or direct clinical decision support.
- Current limitations
- This release includes no structured age, sex, anatomical site, laterality, acquisition-device, visit, diagnosis, treatment or outcome metadata. JSON geometry arrays are empty, and no predefined training, validation or test split is supplied.
- Mask labels
- The seven labels are other, skin, granulation, slough, necrose, tendon and bone. The RGB mask files are the authoritative annotation source.
Access requires all five steps.
Approval is personal, project-specific and non-transferable. Commercial R&D and product-development requests follow the contact route instead of this research application.
- 01Create and verify a personal portal account.
- 02Complete the researcher profile and upload responsible research-data training evidence accepted by Amatis; this may be from CITI or an equivalent programme.
- 03Read and accept the current Data Use Agreement with your legal name.
- 04Submit the research purpose, methods, expected outputs, team, storage, access-control, retention and publication plans.
- 05Receive manual Amatis approval; before transfer, confirm your password and the short code sent by email.
GitHub repository
Review the source code, training workflow and evaluation protocol used for reproducible segmentation comparisons.
Open software and restricted-access data have different boundaries.
Public repository
The Segmentation Benchmarking repository distributes research software, not a dataset. Its code, training workflow and evaluation protocol can be reviewed openly.
Restricted-access dataset
The Amatis wound dataset is provided only to approved researchers under a separate agreement. Commercial product and R&D inquiries follow the contact route.