DIAMS Soil Fauna dataset

Soil Fauna Dataset, containing : 

Raw soil images taken every 6 hours by 4 scanners over a 3 months period, for a total of 610 10200x14032 RGB images. Further information about their acquisition can be found in [1]

Labeled boxes automatically extracted from these raw images, and then manually annotated. These represent about 56k boxes, 15% of which actually contain invertebrates. Invertebrates are classified into 10 different classes. More information can be found in [2], which describes in details the method used to extract boxes, and an automated classification pipeline built using this dataset. 

[1] Emma Belaud, Christophe Jourdan, Dominique Barry-Etienne, Claire Marsden, Agnès Robin, Elisa Taschen,and Mickael Hedde. In situ soil imaging, a tool for monitoring the hourly to monthly temporal dynamicsof soil biota. Biology and Fertility of Soils, 60(8):1055–1071, 2024.

[2] Hadrien Hendrikx, Emma Belaud, François Postic, Morgan Scalabrino, Marion Lebeau, Guerric Le Maire, Christophe Jourdan, Philippe Gallet, Mickael Hedde. Linking automated image analysis to ecological inference: high-throughput monitoring of soil fauna. bioRxiv, 2026