# Salmon and Pollock Object Detection in Commercial Pollock Trawl Video (NOAA AFSC MACE, 2019-2020)

## Facts

- Provider: NOAA Alaska Fisheries Science Center, MACE Program (Wilson et al.)
- License: CC0-1.0 (https://www.fisheries.noaa.gov/inport/item/69412)
- Format: JSON, PNG (in sampled files)
- Stored at: Google Cloud Storage, gs://nmfs_odp_afsc/RACE/MACE/salmon_pollock_object_detection (us)
- Access: public, readable without an account
- DOI: https://doi.org/10.1093/icesjms/fsaf168
- Category: Computer Vision
- Source: https://doi.org/10.1093/icesjms/fsaf168
- Page: https://scigantic.com/archive/c6d16dca-e118-54e5-9e08-44fbd514775b
- Keywords: Computer Vision, JSON, PNG, Google Cloud Storage

## About

Annotated video frames, detection models and raw tow videos used to fine-tune and evaluate object detectors for Pacific salmon (Oncorhynchus spp.) and walleye pollock (Gadus chalcogrammus) in footage from a bycatch reduction device in the Alaska commercial pollock trawl fishery. The annotations comprise 11,572 salmon and 73,394 pollock track annotations over 16,989 frames drawn from 184 video clips, with frame-level labels describing trawl and background conditions; annotated_data/ stores one folder per clip holding PNG frames and a COCO-format JSON. models/ holds the five-fold cross-validated EfficientDet D2, YOLO11 and YOLO12 checkpoints, including a single-class salmon-only YOLO11n, and tow_data/ holds three unannotated fishing tows as MOV videos with the best-model and salmon-only CSV detections. documentation/ explains the COCO layout, data fields and model configurations. Measured 2026-11-15: 17,338 objects, about 74 GB (16,998 PNG frames, 184 JSON, 53 CSV, 52 MOV). Reported performance: the YOLO models beat EfficientDet, detecting on average 90% of fish at 72% accuracy with a 50% IoU threshold. NOAA work released CC0-1.0; cite Wilson et al., ICES Journal of Marine Science, doi 10.1093/icesjms/fsaf168.

## Load the data

```python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('nmfs_odp_afsc/RACE/MACE/salmon_pollock_object_detection/', detail=False)[:10]
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

File types in sampled files: JSON, PNG

Example file: `RACE/MACE/salmon_pollock_object_detection/annotated_data/2019_t15_vid10_4260_4319/2019_t15_vid10_4260_4319.json`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/c6d16dca-e118-54e5-9e08-44fbd514775b
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/c6d16dca-e118-54e5-9e08-44fbd514775b
- Over MCP: `get_schema_card` with id `c6d16dca-e118-54e5-9e08-44fbd514775b` at https://api.scigantic.com/mcp

## Related topics

- [JSON datasets](https://scigantic.com/topics/json)
- [PNG datasets](https://scigantic.com/topics/png)
- [Computer Vision datasets](https://scigantic.com/topics/computer-vision)
