This readme file was generated on 2026-06-29 by Maximilian Neidhardt

# GENERAL INFORMATION

Title of Dataset: 
Zebrafish RGB Camera Tracking
https://doi.org/10.15480/882.17350

Name: Maximilian Neidhardt
ORCID: 0000-0002-5107-0864
Institution: Institute of Medical Technology and Intelligent Systems
Address: Am Schwarzenberg-Campus 1, 21073 Hamburg
Email: Maximilian.Neidhardt@tuhh.de

Name: Debayan Bhattacharya
ORCID: 0000-0001-8552-2227
Institution: Institute of Medical Technology and Intelligent Systems
Address: Am Schwarzenberg-Campus 1, 21073 Hamburg
Email: debayan.bhattacharya@tuhh.de

Date of data collection: 2024-11-01 - 2025-05-01
Geographic location of data collection: Hamburg, Germany
The data collection was partially funded by the Interdisciplinary Competence Center for Interface Research (ICCIR) supported by Hamburg University of Technology (TUHH) and University Medical Center Hamburg-Eppendorf (UKE), TUHH $i^3$ initiative, the RTG 3144 'Multiscale Imaging and Analytics of Interfaces in Musculoskeletal Health' funded by the DFG (Grant 547468385), and the FMTHH funding Programme (Grant 03fmthh2018). Publishing fees are supported by the Open Access Publishing Programme of Hamburg University of Technology (TUHH).


# SHARING/ACCESS INFORMATION
The data is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license and may be used freely, provided that appropriate attribution is given. Thank you!
Neidhardt, Maximilian; Bhattacharya, Debayan (2026): Robust Motion Tracking and Classification of Zebrafish with Deep Learning. IEEE Access. DOI: https://doi.org/10.15480/882.17350

# DATA & FILE OVERVIEW

## File List:
* 2_Fish
    ** 2_Fish_Camera_*.mkv - Recordings of both cameras, 1000 frames each
    ** Annotated_2_Fish_Camera_*.txt - 1000 by hand annotated frames of both cameras
* 5_Fish
    ** 5_Fish_Camera_*.mkv - Recordings of both cameras, 1000 frames each
    ** Annotated_5_Fish_Camera_*.txt - 1000 by hand annotated frames of both cameras
* 10_Fish
    ** 10_Fish_Camera_*.mkv - Recordings of both cameras, 33069 frames each
    ** Annotated_10_Fish_Camera_*.txt - 1000 by hand annotated frames of both cameras
* Training Data
    *  *_Fish - Recordings of both cameras, 5000 frames each, for each individual fish swimming alone in the tank
* Predictions - Tracked bounding boxes with our best performing proposed method ('unprocessed', patchsize 120 x 120 pixel) [Neidhardt, Maximilian, et al. "Robust Motion Tracking and Classification of Zebrafish with Deep Learning." IEEE Access (2026)]
* Models - YOLO v12 Weights of the detection & classification network to process images of camera 0 and 1, respectively
* VideoSubmission.mov - Video showing our tracker's output
* camera_parameters.txt - OpenCV estimated camera parameters

# METHODOLOGICAL INFORMATION

## Description of methods used for collection/generation of data: 
Two RGB cameras (Basler, acA4112-20um) are mounted at the top of a fish tank in parallel. Synchronized image data acquisition is enabled via a trigger. We estimate all camera parameters with Zhang's camera calibration algorithm provided by OpenCV (version 4.10.0). All camera recordings were performed at 20 fps with an image resolution of 4096 × 3000 pixels (4K). To extract PNG images from the encoded video data, we recommend using the *FFmpeg* package. Images can be generated with:

```bash
ffmpeg -threads 8 -i <input_video.264> <output_directory>/frame_%05d.png
```
In order to evalaute a tracker performance we recommend using the py-motmetrics repository. All datasets are reported in the MOT16 format.

https://github.com/cheind/py-motmetrics

A detailed discription of the experimental setup and the acquired data can be found in Neidhardt, Maximilian, et al. "Robust Motion Tracking and Classification of Zebrafish with Deep Learning." IEEE Access (2026).