Datasets:

EcodylicScience
/

mosaic-example-zebrafish

---
license: cc0-1.0
language:
  - en
tags:
  - biology
  - animal-behavior
  - collective-behavior
  - animal-tracking
  - mosaic
pretty_name: "Zebrafish (wild type): collective motion and social forces"
size_categories:
  - 10M<n<100M
---
# Zebrafish wild type, collective motion and social forces

Pose tracks for 33 trials of wild-type (AB strain) zebrafish, six per group,
free-swimming in a circular arena about 48 cm across, filmed from above at 60 Hz
for 30 minutes. Converted to [mosaic](https://github.com/EcodylicScience/mosaic)
parquet tables and ready to analyse without a conversion step.

Worked example:
[`notebooks/collective-motion-zebrafish.ipynb`](https://github.com/EcodylicScience/mosaic/blob/main/notebooks/collective-motion-zebrafish.ipynb).
Set `SOURCE = "download"` (the default) and it fetches both files below.

## A tracker with no keypoints

This is a Ctrax/JAABA-style `trx` tracker: it reports a centroid, an orientation
and a fitted ellipse, and **locates no keypoints at all**. So these tables carry
`X`, `Y` and `ANGLE` and **no `poseX*` / `poseY*` columns** — which makes the
dataset a useful case for anyone whose own tracker does the same. Features that
genuinely need keypoints refuse a table without them, naming themselves; features
that merely prefer them fall back to the body centre.

`X` / `Y` is the tracker's **centroid**, as `mosaic_v1` asks. (Contrast the
[shiners dataset](https://huggingface.co/datasets/EcodylicScience/mosaic-example-shiners),
where it is the head.)

## Files

| File | Size | What it is |
| --- | --- | --- |
| `zebrafish-wt-tracks.tar.gz` | 767 MB | A complete mosaic dataset: `dataset.yaml` + `tracks/`. 33 trials, about 21.4 million rows. |
| `zebrafish-raw-sample.tar.gz` | 79 MB | Two trials of raw `.mat`, copied verbatim. |

`SHA256SUMS` carries the checksums.

The raw sample exists so the example notebook can *run* its converter rather than
only display it. Converting the two samples reproduces the shipped tables exactly,
and the notebook asserts this before going on.

## Contents

Columns: `frame, time, id, group, sequence, X, Y, ANGLE, body_length,
body_width, px_per_cm`. Positions are **video pixels**; `px_per_cm` is 40.236 and
is carried per row so the calibration travels with the data, but the columns are
not scaled by it — a physical unit is obtained downstream by mosaic's
`scale-to-cm` feature, never stored in the table.

`ANGLE` is the ellipse fit's head-resolved orientation, which the tracker
measures directly — hence the `zfish_trx_v1` schema (`mosaic_v1` plus `ANGLE`)
rather than plain `mosaic_v1`.

`group` is empty for every row: these trials form one flat set with no grouping
level. That is legal — `group` is an optional namespace, and `sequence` alone
identifies each trial.

## Use

```python
import tarfile
from huggingface_hub import hf_hub_download
from mosaic.core.dataset import Dataset
path = hf_hub_download("EcodylicScience/mosaic-example-zebrafish",
                       "zebrafish-wt-tracks.tar.gz", repo_type="dataset")
with tarfile.open(path) as tar:
    tar.extractall("data")
ds = Dataset("data/zebrafish-wt-tracks/dataset.yaml").load()
tracks = ds.load_tracks("", "data10282015_10_43_AMAB")
```

`huggingface_hub >= 1.2.0` is worth pinning: older clients retry a rate-limit
response with a 25-second backoff against a 5-minute window, so they fail where a
current client waits.

## Provenance and changes

Derived from the Dryad record accompanying Tang et al. (2020). **CC0 1.0**, so
this derivative carries the same terms.

Source: [doi:10.5061/dryad.hx3ffbg9n](https://doi.org/10.5061/dryad.hx3ffbg9n)

What was changed:

- The `trx` MATLAB struct arrays were converted to mosaic `zfish_trx_v1` parquet
  tables. Column names and layout differ from the original release. **Positions
  and orientations are unaltered** — `x`, `y` and `theta` are carried through
  verbatim.
- **A failed ellipse fit is reported as an exact zero in the source, and is
  converted to NaN here.** This affects 12% to 46% of rows depending on the
  trial, 21% overall. Read literally, a zero is a fish zero pixels long: it drags
  the median body length down and would collapse any derived pose onto the
  centroid. This is the one place a value is changed rather than carried across,
  and it is the reason to prefer these tables over a naive read of the source.
- Fields not converted: `arena` (stale fly-bowl template data — it reports a
  radius of 254 px while the fish occupy 108–1983 px), and the movie-name and
  bookkeeping fields.

Not included: video. Only pose data is distributed in the source record.

## Citation

Tang W, Davidson JD, Zhang G, Conen KE, Fang J, Serluca F, Li J, Xiong X, Coble
M, Tsai T, Molind G, Fawcett CH, Sanchez E, Zhu P, Couzin ID, Fishman MC (2020)
Genetic Control of Collective Behavior in Zebrafish. *iScience* 23(3): 100942.
[doi:10.1016/j.isci.2020.100942](https://doi.org/10.1016/j.isci.2020.100942)

	---
	license: cc0-1.0
	language:
	- en
	tags:
	- biology
	- animal-behavior
	- collective-behavior
	- animal-tracking
	- mosaic
	pretty_name: "Zebrafish (wild type): collective motion and social forces"
	size_categories:
	- 10M<n<100M
	---

	# Zebrafish wild type, collective motion and social forces

	Pose tracks for 33 trials of wild-type (AB strain) zebrafish, six per group,
	free-swimming in a circular arena about 48 cm across, filmed from above at 60 Hz
	for 30 minutes. Converted to [mosaic](https://github.com/EcodylicScience/mosaic)
	parquet tables and ready to analyse without a conversion step.

	Worked example:
	[`notebooks/collective-motion-zebrafish.ipynb`](https://github.com/EcodylicScience/mosaic/blob/main/notebooks/collective-motion-zebrafish.ipynb).
	Set `SOURCE = "download"` (the default) and it fetches both files below.

	## A tracker with no keypoints

	This is a Ctrax/JAABA-style `trx` tracker: it reports a centroid, an orientation
	and a fitted ellipse, and locates no keypoints at all. So these tables carry
	`X`, `Y` and `ANGLE` and *no `poseX` / `poseY` columns* — which makes the
	dataset a useful case for anyone whose own tracker does the same. Features that
	genuinely need keypoints refuse a table without them, naming themselves; features
	that merely prefer them fall back to the body centre.

	`X` / `Y` is the tracker's centroid, as `mosaic_v1` asks. (Contrast the
	[shiners dataset](https://huggingface.co/datasets/EcodylicScience/mosaic-example-shiners),
	where it is the head.)

	## Files

	\| File \| Size \| What it is \|
	\| --- \| --- \| --- \|
	\| `zebrafish-wt-tracks.tar.gz` \| 767 MB \| A complete mosaic dataset: `dataset.yaml` + `tracks/`. 33 trials, about 21.4 million rows. \|
	\| `zebrafish-raw-sample.tar.gz` \| 79 MB \| Two trials of raw `.mat`, copied verbatim. \|

	`SHA256SUMS` carries the checksums.

	The raw sample exists so the example notebook can run its converter rather than
	only display it. Converting the two samples reproduces the shipped tables exactly,
	and the notebook asserts this before going on.

	## Contents

	Columns: `frame, time, id, group, sequence, X, Y, ANGLE, body_length,
	body_width, px_per_cm`. Positions are video pixels; `px_per_cm` is 40.236 and
	is carried per row so the calibration travels with the data, but the columns are
	not scaled by it — a physical unit is obtained downstream by mosaic's
	`scale-to-cm` feature, never stored in the table.

	`ANGLE` is the ellipse fit's head-resolved orientation, which the tracker
	measures directly — hence the `zfish_trx_v1` schema (`mosaic_v1` plus `ANGLE`)
	rather than plain `mosaic_v1`.

	`group` is empty for every row: these trials form one flat set with no grouping
	level. That is legal — `group` is an optional namespace, and `sequence` alone
	identifies each trial.

	## Use

	```python
	import tarfile
	from huggingface_hub import hf_hub_download
	from mosaic.core.dataset import Dataset

	path = hf_hub_download("EcodylicScience/mosaic-example-zebrafish",
	"zebrafish-wt-tracks.tar.gz", repo_type="dataset")
	with tarfile.open(path) as tar:
	tar.extractall("data")

	ds = Dataset("data/zebrafish-wt-tracks/dataset.yaml").load()
	tracks = ds.load_tracks("", "data10282015_10_43_AMAB")
	```

	`huggingface_hub >= 1.2.0` is worth pinning: older clients retry a rate-limit
	response with a 25-second backoff against a 5-minute window, so they fail where a
	current client waits.

	## Provenance and changes

	Derived from the Dryad record accompanying Tang et al. (2020). CC0 1.0, so
	this derivative carries the same terms.

	Source: [doi:10.5061/dryad.hx3ffbg9n](https://doi.org/10.5061/dryad.hx3ffbg9n)

	What was changed:

	- The `trx` MATLAB struct arrays were converted to mosaic `zfish_trx_v1` parquet
	tables. Column names and layout differ from the original release. **Positions
	and orientations are unaltered** — `x`, `y` and `theta` are carried through
	verbatim.
	- **A failed ellipse fit is reported as an exact zero in the source, and is
	converted to NaN here.** This affects 12% to 46% of rows depending on the
	trial, 21% overall. Read literally, a zero is a fish zero pixels long: it drags
	the median body length down and would collapse any derived pose onto the
	centroid. This is the one place a value is changed rather than carried across,
	and it is the reason to prefer these tables over a naive read of the source.
	- Fields not converted: `arena` (stale fly-bowl template data — it reports a
	radius of 254 px while the fish occupy 108–1983 px), and the movie-name and
	bookkeeping fields.

	Not included: video. Only pose data is distributed in the source record.

	## Citation

	Tang W, Davidson JD, Zhang G, Conen KE, Fang J, Serluca F, Li J, Xiong X, Coble
	M, Tsai T, Molind G, Fawcett CH, Sanchez E, Zhu P, Couzin ID, Fishman MC (2020)
	Genetic Control of Collective Behavior in Zebrafish. iScience 23(3): 100942.
	[doi:10.1016/j.isci.2020.100942](https://doi.org/10.1016/j.isci.2020.100942)