mikkoim/aquamonitor-jyu

AquaMonitor JYU

This is a sample dataset of aquatic invertebrates from 31 classes. The training split has 40,880 images from 1049 individuals, while the validation split has 6394 images from 157 individuals.

Metadata columns

Column Name	Type	Description
img	str	The image filename. Unique identifier across the dataset.
individual	str	Identifier for the individual specimens.
imaging_run	str	Identifier for the imaging run. Some specimens have been imaged multiple times, and this ID refers to a single "drop" through the cuvette. Consists of individual id with running number suffix. With bulk imaged samples each individual has been imaged only once, resulting in 'individual' == 'imaging_run'.
taxon	str	The scientific name of the specimen. Refers to both adult and juvenile forms.
taxon_group	str	Specifies the visual group of the specimen. Some taxa have specimens from multiple visual groups, like juvenile and adult forms. If the taxon has no such subgroups, this value is the same as 'taxon'. Used for classification benchmarks.
taxon_label	str	Original label used during morphological identification. Matches the labels used in the monitoring databases. Was used during imaging and matches image filenames.
taxon_code	str	The abbreviation of the taxon_label. Matches filenames and identifiers.
lake	str	The lake the specimen was collected from.
site	str	Sampling site.
sample	str	The taxon-site combination. Before imaging, the samples were stored in containers with these groupings.
plate	str	(Only on DNA and biomass subsets) The 96 well plate ID.
position	str	(Only on DNA and biomass subsets) The position on the 96 well plate. Values range in [A-H] for rows and [1-12] for columns.
camera	int	The camera the image is from. Values in {1,2}.
area	int	The area of the specimen in the image. From BIODISCOVER imaging software.
perimeter	int	The perimeter of the specimen in the image. From BIODISCOVER imaging software.
holes	int	Number of morphological holes in the image segmentation. From BIODISCOVER imaging software.
max_feret_diameter	int	Maximum Feret diameter of the segmentation object. From BIODISCOVER imaging software.
area_holes	int	The area of the morphological holes in the segmentation. From BIODISCOVER imaging software.
roi_left	int	The left position of the image crop, relational to the full cuvette.
roi_top	int	The top position of the image crop, relational to the full cuvette.
roi_right	int	The right position of the image crop, relational to the full cuvette.
roi_bottom	int	The bottom position of the image crop, relational to the full cuvette.
width	int	Width of the image.
height	int	Height of the image.
imaging_time	datetime	Imaging timestamp.
dataset	str	The dataset subset image belongs to. Values in {'dna', 'biomass', 'bulk21', 'bulk22'}.
year	str	Sampling year. Values in {'2021', '2022'}.
is_bulk	bool	Whether the image was taken in the bulk imaging phase.
has_dna	bool	The individual has DNA sequence data.
has_biomass	bool	The individual has biomass information.
fold0-fold4	str	The benchmark split the image belongs to, for five cross-validation folds. Values in {'train', 'val', 'test', '0'}. Values of '0' correspond to images in classes that have less than 50 specimens.
well_col	str	Column in the value 'position'. Values in range [1-12].
well_row	str	Row in the value 'position'. Values in range [A-H].
weight	float	Weight of the specimen in milligrams. Calculated from plate_weight and total_weight.
log_weight	float	Natural log of the weight.
log_weight_xplus1	float	Weight in the form ln(x+1), where x is the value in 'weight'.
plate_weight	float	The measured plate weight.
total_weight	float	Weight of the plate with the specimen.
caddisfly_without_shell	float	If value is not NaN, this is the total weight of caddisfly samples weighed without the shell.