Healthy and Loser Salmon Dataset

The "Healthy and Loser Salmon Dataset" contains images of farmed Atlantic salmon in sea cages, and corresponding annotations. For the detailed description of the dataset, the user is referred to the published open-access article "Identifying losers: Automatic identification of growth-stunted salmon in aquaculture using computer vision", which can be downloaded at https://doi.org/10.1016/j.mlwa.2024.100562.

The dataset contains a total of 207 images and their respective annotations files ("json" and "yolo" formats) for the bounding boxes of the 2 classes to be detected (healthy and loser salmon).
These 207 images present a total of 1750 instances of fish (1319 healthy and 431 loser).

The dataset is divided into three independent collections: training, validation and test sets.
The training set comprises 145 images, 888 instances of healthy fish and 306 loser salmons;
while the validation set presents 41 images, 260 instances of healthy fish and 86 of loser salmons;
and the test set has 21 images, 171 healthy fish instances and 39 loser instances.

The directories are organized as follows:

healthy_loser_salmon_dataset
json
test
train
valid
yolo
data.yaml
images
test
train
valid
labels
test
train
valid

The "json" directory presents three folders: "test", "train" and "valid". Each one of these directories presents the images (jpg) that belong to the respective set and their following json files (created using the "LabelMe" software) with the annotation of the bounding boxes.

The "yolo" directory has the classic structure used by Yolo, with a "data.yaml" file defining each set, number of classes and name of each class; the "image" folder that contains the images (png) of each set and is divided as test, train and valid, referring to each set; and finally, the directory "label", that presents the txt files of each location of the bounding boxes on the yolo pattern.