Description

Technical info

Data preprocessing

The following pre-processing was applied to each image on the Roboflow platform:

• Auto-orientation of pixel data (with EXIF-orientation stripping
• Resize to 640x640 (Stretch)

The following augmentation was applied to create 3 versions of each source image: 

• Random brigthness adjustment of between -25 and +25 percent
• Random exposure adjustment of between -23 and +23 percent
• Random Gaussian blur of between 0 and 3 pixels
• Salt and pepper noise was applied to 1.88 percent of pixels

Data splitting

The initial training dataset was split as follows: 86% Training Set, 10%  Validation Set and 4% Test Set.

Data labelling

The classes used in the initial dataset are:

Eel, Fish, Flat Fish, Jelly, Pipefish, Ray, Seahorse, Shark

Parameters

NA

Data sources

The imagery used in this training dataset consists of image frame captures from the Smartbay video Archive files, CC-BY imagery from the www.minka-sdg.org website and images taken by Eva Cullen in the "Galway Atlantaquaria" Aquarium in Galway, Ireland.

Data quality

Video frames were extracted from Smart Observatory and Galway Atlantaquaria video recordings for annotation usign CVAT and also ffmpeg.

Target species CC-BY images were also downloaded from minka-sdg.org using a python minka-downloader script (https://github.com/obsea-upc/minka-downloader)

Useful images were manually selected and annotated by the bursor students and collated into training datasets in Roboflow.  

Data resizing

The images in the initial dataset were resized to 640x640 when exported from Roboflow.

Spatial coverage

Smartbay Observatory Spiddal, Galway Bay, Ireland

Galway Atlantaquaria, Seapoint Promenade Galway H91 T2FD, Ireland

Minka-sdg.org Sample Imagery from Spanish and Portuguese waters 

Contact information

Data.Requests@Marine.ie