---
license: cc0-1.0
language:
  - en
tags:
  - biology
  - animal-behavior
  - collective-behavior
  - animal-tracking
  - object-detection
  - mosaic
pretty_name: "Golden shiners: point-model training and TREx tracking"
size_categories:
  - 1M<n<10M
---
# Golden shiners — POLO point-model training and TREx tracking example

A [mosaic](https://github.com/EcodylicScience/mosaic) dataset for a complete
tracking workflow: published tracks become point pseudo-annotations, those train a
POLO point detector, and TREx tracks with it — or, with no model at all, with its own
background subtraction.

Worked end to end in
[`notebooks/shiners-polo-tracking.ipynb`](https://github.com/EcodylicScience/mosaic/blob/main/notebooks/shiners-polo-tracking.ipynb).

## What is in it

| File | Size | Contents |
| --- | --- | --- |
| `shiners-polo.tar.gz` | 1.09 GB | `dataset.yaml`, `media_raw/` (3 concatenated recordings), `media/frames/kmeans/` (80 frames per recording), `tracks/` (the published SchoolTracker tables) |

The POLO training tree is not shipped: it is a rearrangement of the frames and the
tables under different filenames, so it would enlarge the download to carry nothing
new. Section 2 of the notebook rebuilds it in seconds.

The tables under `tracks/` were produced by a converter that lives in
[`collective-motion-shiners.ipynb`](https://github.com/EcodylicScience/mosaic/blob/main/notebooks/collective-motion-shiners.ipynb),
not inside mosaic, so their `src_format` is one no installed mosaic knows. Reading
them works; re-converting from source needs that notebook's converter cell.

## Contents

`shiners-polo.tar.gz` unpacks to `shiners-polo/`, a mosaic dataset:

| Root | What |
| --- | --- |
| `media_raw/` | 3 re-encoded H.264 recordings, one per group size (10 / 30 / 70 fish), ~24,600 frames each |
| `media/frames/` | 80 k-means-selected PNG frames per recording (240 total) |
| `tracks/` | the published SchoolTracker tables, identity-corrected |
| `dataset.yaml` | the manifest |

```bash
tar xzf shiners-polo.tar.gz
```

Then point the notebook's `LOCAL_DATASET` at `shiners-polo/`, or let it download this
archive itself.

## No model weights

None are included. The trainer, Ultralytics, is AGPL-3.0; rather than work through what
distributing a model produced by it requires, this dataset carries none. The notebook's
model-free TREx route needs no weights, and its training section prints the command to
produce your own.

## Provenance and licence

Free-swimming golden shiners (*Notemigonus crysoleucas*) in a 2.1 x 1.2 m tank, filmed
from above, tracked with SchoolTracker and manually identity-corrected.

Source data CC0 1.0 from Dryad,
[doi:10.5061/dryad.sbcc2fr2h](https://doi.org/10.5061/dryad.sbcc2fr2h). The recordings
here are re-encoded (each trial's segments concatenated, first frame of each segment
dropped to match the track tables) and the tables converted to mosaic's schema; both are
derivatives of that record and carry the same licence.

**Please cite the original work:**

> Davidson JD, Sosna MMG, Twomey CR, Sridhar VH, Leblanc SP, Couzin ID (2021)
> Collective detection based on visual information in animal groups.
> *Journal of the Royal Society Interface* 18: 20210142.
> [doi:10.1098/rsif.2021.0142](https://doi.org/10.1098/rsif.2021.0142)

## A note on `X`/`Y`

In these tables `X`/`Y` is the fish's **head**, not its body centre — SchoolTracker
reports it that way. It matters when training a point model on them, and the notebook
says why.