# Third-Party Sources

Nereus is a collection rather than a single-license dataset. This document
records known upstream sources and the boundaries between upstream material and
Nereus-authored additions.

## Source Matrix

| Nereus path | Upstream source | Upstream material | License status |
| --- | --- | --- | --- |
| `grounded_counting/images/` | IOCfish5K | Original underwater images | CC BY-NC 4.0 |
| `grounded_counting/annotations/final_annotations/` | IOCfish5K plus Nereus | Original point annotations and Nereus task enrichments | IOCfish5K content: CC BY-NC 4.0; Nereus additions: root LICENSE |
| `grounded_counting/metadata/splits/iocfish5k_original/` | IOCfish5K | Original train, validation, and test manifests | CC BY-NC 4.0 |
| `grounded_counting/qa/` | Nereus, derived from IOCfish5K | Region, camouflage, density/depth, and ecological-reasoning QA | Root LICENSE, subject to IOCfish5K source terms |
| `object_attribute_understanding/` | FathomNet and its contributors; other sources may require per-item verification | Marine images, bounding boxes, concepts, and derived attributes | Per-item source license; no repository-wide license |

## IOCfish5K

- Dataset: IOCfish5K and IOCfish5K-D
- Project: Indiscernible Object Counting in Underwater Scenes
- Authors: Guolei Sun, Zhaochong An, Yun Liu, Ce Liu, Christos Sakaridis,
  Deng-Ping Fan, and Luc Van Gool
- Venue: IEEE/CVF CVPR 2023
- Official repository:
  https://github.com/GuoleiSun/Indiscernible-Object-Counting
- Paper:
  https://openaccess.thecvf.com/content/CVPR2023/html/Sun_Indiscernible_Object_Counting_in_Underwater_Scenes_CVPR_2023_paper.html
- License: Creative Commons Attribution-NonCommercial 4.0 International
  (CC BY-NC 4.0)
- License text:
  https://creativecommons.org/licenses/by-nc/4.0/

The `grounded_counting` component includes 5,637 IOCfish5K images and aligned
point-based annotations. Nereus retains four-digit release IDs while preserving
source-annotation metadata needed for traceability. Byte-for-byte copies of
the original IOCfish5K `train_id.txt`, `val_id.txt`, and `test_id.txt`
manifests are retained under
`grounded_counting/metadata/splits/iocfish5k_original/`.

Required attribution includes the IOCfish5K project and paper. Commercial use
is not permitted by CC BY-NC 4.0 without separate permission from the
IOCfish5K rights holders.

## FathomNet

- Project: FathomNet
- Operator: Monterey Bay Aquarium Research Institute (MBARI)
- Database: https://database.fathomnet.org/
- Project site: https://www.fathomnet.org/
- Data use policy: https://www.fathomnet.org/datause
- Terms of use: https://www.fathomnet.org/terms
- Reference paper:
  https://doi.org/10.1038/s41598-022-19939-2

FathomNet's current policy allows visual-content contributors to select from
CC0, CC BY, CC BY-NC, or CC BY-NC-ND when sharing content outside the Portal.
The applicable owner retains copyright. FathomNet annotations may also carry
source-specific terms.

The FathomNet image API records inspected during this release audit expose UUID,
URL, contributor, source tag, dimensions, SHA-256, and bounding boxes, but do
not expose a license field. Consequently, the existence of a matching
FathomNet UUID or public URL must not be interpreted as a blanket license.

This grounded-counting integration does not add or re-upload FathomNet images.
A future object-attribute release should include a per-item provenance manifest
containing, at minimum:

- Nereus file path
- FathomNet UUID
- source URL
- source organization or owner
- required attribution
- exact license identifier
- source metadata retrieval date
- SHA-256

Items without verifiable redistribution terms should not be added to a new
public release.

## Nereus-Authored Additions

Nereus-authored material includes task-specific enrichments, QA data, canonical
metadata, split manifests, and release documentation. Its license is defined in
the root `LICENSE`, but that license does not supersede upstream terms.