SPARUS-LD:
Gilthead Seabream (Sparus aurata) Landmark Detection

Igor Talijančić1, Josip Šarić2,3, Josip Zavada3, Luka Žuvić1, Siniša Šegvić3, Tanja Šegvić-Bubić1
1Institute of Oceanography and Fisheries
2University of Zagreb, Faculty of Electrical Engineering and Computing
3University of Ljubljana, Faculty of Computer and Information Science
Code Dataset

Our method detects 18 different landmarks of the gilthead seabream (Sparus aurata).

Abstract

The SPARUS-LD dataset is the input for geometric morphometric analysis of gilthead seabream body shape and serves as a valuable resource for the assessment of measurement error in landmark digitization. To address the subjectivity and inconsistencies in manual annotation, a deep learning model was developed to standardize the process. The performance of the model was evaluated against human operators with different levels of experience by comparing automated and manually digitized landmarks in a test dataset. This approach not only quantifies the measurement error, but also examines how the operators experience affects its magnitude. The machine learning model provides an objective and repeatable method that minimizes human error and reduces variability in digitization. The SPARUS-LD dataset highlights the challenges posed by morphological variation in fish species and demonstrates the importance of validating automated methods against human performance. It serves as an essential resource for assessing the reliability and accuracy of both manual and automated landmark placement in the context of geometric morphometrics and contributes to the ongoing development of more robust morphometric analysis methods.

Dataset description

SPARUS-LD dataset contains 2052 high-resolution images of gilthead seabream (Sparus aurata) specimens annotated with 18 landmarks/keypoints. Additionally, each specimen is associated with one of the three classes according to its origin: wild, farmed, or farm-associated.

We split the dataset into training, validation and test sets. The table below shows the distribution of origin in each set.

Origin Training Validation Test
Wild 419 97 190
Farmed 411 100 324
Farm-associated 254 51 206
Total 1084 248 720

Landmarks are defined as follows:
  1. anterior tip of snout at the upper jaw
  2. vertical point above the most anterior point in the eye
  3. anterior insertion of the dorsal fin
  4. last spiny ray of the dorsal fin
  5. posterior insertion of the dorsal fin
  6. dorsal point at the least depth of the caudal peduncle
  7. posterior body extremity
  8. ventral point at the least depth of the caudal peduncle
  9. posterior insertion of the anal fin
  10. anterior insertion of the anal fin
  1. insertion of the pelvic fin
  2. ventral tip of the insertion of the operculum on the lateral profile
  3. point of maximum extension of the operculum on the lateral profile
  4. anterior extremity of the lateral line on the head profile
  5. dorsal insertion of the pectoral fin
  6. ventral insertion of the pectoral fin
  7. the most anterior point in the eye
  8. the most posterior point in the eye

Below you can see a couple of examples from our dataset. Hover over the keypoint for description.

Dataset download

The dataset is available at Zenodo.
For more information, please contact us via e-mail.

BibTeX

If you use our dataset in your research, please cite the following: 


@misc{sparus-ld,
  title={SPARUS-LD: Gilthead Seabream (Sparus aurata) Landmark Detection},
  author={Igor Talijančić and Josip Šarić and Josip Zavada and Luka Žuvić and Siniša Šegvić and Tanja Šegvić-Bubić},
  year={2024},
}

Acknowledgments

This research has been fully supported by Croatian Science Foundation under the project IP-2022-10-7232 Enhancing Environmental Performance of Net-Pen Marine Aquaculture. Josip Šarić is supported by the European Union's Horizon Europe research and innovation programme under the Marie Skłodowska-Curie Postdoctoral Fellowship Programme, SMASH co-funded under the grant agreement No. 101081355. The SMASH project is co-funded by the Republic of Slovenia and the European Union from the European Regional Development Fund.