Synthetic images of corals (Desmophyllum pertusum) with object detection models

SND-ID: 2022-98-1. Version: 1. DOI: https://doi.org/10.5878/hp35-4809

Citation

Creator/Principal investigator(s)

Matthias Obst - University of Gothenburg, Department of Marine Sciences orcid

Sarah Al-Khateeb - MMT Sweden AB / Ocean Infinity

Victor Anton - wildlife.ai orcid

Jannes Germishuys - Combine AB

Research principal

University of Gothenburg - Department of Marine Sciences rorId

Description

Two object detection models using Darknet/YOLOv4 were trained on images of the coral Desmophyllum pertusum from the Kosterhavet National Park. In one of the models, the training image data was amplified using StyleGAN2 generative modeling.
The dataset contains 2266 synthetic images with labels and 409 original images of corals used for training the ML model. Included is also the YOLOv4 models and the StyleGAN2 network.

The still images were extracted from raw video data collected using a remotely operated underwater vehicle.
409 JPEG images from the raw video data are provided in 720x576 resolution. In certain images, coordinates visible in the OSD have been cropped.
The synthetic images are PNG files in 512x512 resolution.
The StyleGAN2 network is included as a serialized pickle file (*.pkl).
The object detection models are provided in the .weights format used by the Darknet/YOLOv4 package. Two files are included (trained on original images only, trained on original + synthetic images).

The machine learning software packages used is currently (2022) available on Github:

... Show more..
Two object detection models using Darknet/YOLOv4 were trained on images of the coral Desmophyllum pertusum from the Kosterhavet National Park. In one of the models, the training image data was amplified using StyleGAN2 generative modeling.
The dataset contains 2266 synthetic images with labels and 409 original images of corals used for training the ML model. Included is also the YOLOv4 models and the StyleGAN2 network.

The still images were extracted from raw video data collected using a remotely operated underwater vehicle.
409 JPEG images from the raw video data are provided in 720x576 resolution. In certain images, coordinates visible in the OSD have been cropped.
The synthetic images are PNG files in 512x512 resolution.
The StyleGAN2 network is included as a serialized pickle file (*.pkl).
The object detection models are provided in the .weights format used by the Darknet/YOLOv4 package. Two files are included (trained on original images only, trained on original + synthetic images).

The machine learning software packages used is currently (2022) available on Github:
StyleGAN2: https://github.com/NVlabs/stylegan2
YOLOv4: https://github.com/AlexeyAB/darknet Show less..

Data contains personal data

No

Language

Method and outcome

Time period(s) investigated

1999 – 2001

Data format / data structure

Data collection

Data collection 1

  • Mode of collection: Recording
  • Description of the mode of collection: Video recordings from 35 research cruises in the Kosterhavet National Park using a ROV.
  • Time period(s) for data collection: 1999 – 2004
  • Data collector: Department of Marine Sciences, University of Gothenburg

Data collection 2

  • Mode of collection: Transcription
  • Description of the mode of collection: The classification of Desmophyllum pertusum in still images from the video data has been performed as citizen science by volunteers using the classification tool on the Koster seafloor observatory website.
  • Data collector: The Koster seafloor observatory
Geographic coverage

Geographic spread

Geographic location: Sweden

Geographic description: Kosterhavet National Park

Administrative information

Responsible department/unit

Department of Marine Sciences

Funding 1

  • Funding agency: Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (FORMAS) rorId
  • Funding agency's reference number: 2021-02465_Formas
  • Project name on the application: National implementation of a platform for analysis of sub-sea images (PLAN-SUBSIM)
  • Funding information: The data set collection was funded though Swedish Biodiversity Data Infrastructure (SRC), Ocean Data Factory (Vinnova), and PLAN-SUBSIM (FORMAS)

Funding 2

  • Funding agency: Swedish Research Council rorId
  • Funding agency's reference number: 2019-00242
  • Project name on the application: Swedish Biodiversity Data Infrastructure
  • Funding information: The data set collection was funded though Swedish Biodiversity Data Infrastructure (SRC), Ocean Data Factory (Vinnova), and PLAN-SUBSIM (FORMAS)

Funding 3

  • Funding agency: Vinnova rorId
  • Funding agency's reference number: 2019-02256
  • Project name on the application: Ocean Data Factory
  • Funding information: The data set collection was funded though Swedish Biodiversity Data Infrastructure (SRC), Ocean Data Factory (Vinnova), and PLAN-SUBSIM (FORMAS)
Topic and keywords

Research area

Zoology (Standard för svensk indelning av forskningsämnen 2011)

Ecology (Standard för svensk indelning av forskningsämnen 2011)

Imagery / base maps / earth cover (INSPIRE topic categories)

Biota (INSPIRE topic categories)

Oceans (INSPIRE topic categories)

Publications

Alkhateeb, Sarah, Obst, Matthias, Anton, Victor and Germishuys Jannes. (2023). A methodology to detect deepwater corals using Generative Adversarial Networks. GigaScience. [Submitted manuscript].

If you have published anything based on these data, please notify us with a reference to your publication(s). If you are responsible for the catalogue entry, you can update the metadata/data description in DORIS.

Published: 2023-04-12
Last updated: 2023-05-03