InsectDCT: Datasets for training and evaluating pipeline processing insect camera recordings

Contains the dataset for training and testing of models to detect and hierarchically classify taxa of primary arthropods in video or time-lapse camera recordings described in the paper:

Kim Bjerge, Simon Wogram, Pau Enric Serra-Marin, Toke T. Høye, "InsectDCT: A generalised pipeline for detection, taxonomic classification, and tracking of insects in camera-trap recordings", 2026, bioRxiv, https://doi.org/10.64898/2026.07.07.736939

The zip files contain the needed files and directory structure to train the models in Python code published at: https://github.com/kimbjerge/insectDCT

=================================================================
Content of ZIP files for the training dataset (insect detection):
=================================================================

detectorColorYOLOV6.zip (51GB) and detectorMotionYOLOV6.zip (71GB)

The ZIP files can be downloaded from: https://anon.erda.au.dk/cgi-sid/ls.py?share_id=AO4fCuU3fi

Contains annotated images and labels in YOLO format: https://github.com/ultralytics/yolov5/issues/2293

The subfolders insects5Motion (Motion-enhanced images) and insects5Color (RGB images) contain:

train/images: *jpg - images for training
train/labels: *txt - YOLO format labels for training
val/images: *jpg - images for validation
val/labels: *txt - YOLO format labels for validation

The images are collected from different projects, with each camera monitoring different backgrounds of vegetation:

GreenRoof: AC-S?*.jpg - (1-5) 5 systems with 10 cameras
GreenHouse: S?_[0|1]*.jpg - (1-4) (0-1) 4 systems with 8 cameras
Arthropods: *NAIM??*.jpg - (01-60) 48 cameras
Arthropods: *BOR??*.jpg - (01-05) 5 cameras
PollWatch: PW_S?_*.jpg - (2,3,4) 5 systems with 10 cameras
MAMBO: MB_*.jpg - (au, cirad, ecoinn, ufz) four countries (DK, FR, MT, DE) 10 cameras
Orchard: OC_pi*.jpg - (8,10,12,15,20,26,28) six cameras

The subfolders insects6Motion (Motion-enhanced images) and insects6Color (RGB images) contain:

trainACSHQ/images: First iteration of selecting insect images in the ASCHQ recorded video clips
trainACSHQ/labels
trainACSHQ2/images: Second iteration of selecting insect images in the ASCHQ recorded video clips
trainACSHQ2/labels

ACSHQ: PP_S??_*.jpg - (0-61) >25 different plants

YOLO_yaml: contains the YAML files for training YOLO11.

============================================================
Content of ZIP file for the test dataset (insect detection):
============================================================

detectorTestV6.zip (35GB)

The ZIP file can be downloaded from: https://anon.erda.au.dk/cgi-sid/ls.py?share_id=AO4fCuU3fi

Color images and Motion-enhanced images (m):

The images are collected from different projects, with each camera monitoring different backgrounds of vegetation:

GreenHouse: testGreenH(m): GH_S?*.jpg - 8 cameras
Heather:  testHeather(m): HE_*.jpg - 25 cameras
MAMBO: testMBO(m): MB_*.jpg - (ukceh, uva) MAMBO two countries (UK, NL) 5 cameras
Orchard: testOrchard(m): OC_pi*.jpg - (10,11,12,13,16,17) 6 cameras with different dates from training data
PollWatch: testPollW(m): PW_S?_*.jpg - 5 systems with 10 cameras
RT-Impact: testRTNI(m): RT_*.jpg - one camera
ASCHQ: testACSHQ(m): PP_S??_*.jpg - >25 different plants

YOLO_yaml: contains the yaml files for testing and validating YOLO11.

======================================================================
Content of ZIP files for the training dataset (insect classification):
======================================================================

classifierTrainV6.zip (4.5GB)

GreenRoof: NI - 11 taxonomic groups
Arthropod: NI2 - 69 taxonomic groups
MAMBO: NI2_MAMBO - 16 taxonomic groups
Orchard: Orchard - 59 taxonomic groups
Supplemented images: GBIF - 52 taxonomic groups of cropped insect images

In total 80 taxonomic groups are present in the dataset see details: classifierDatasetV6.txt

Every subfolder is named, where space separates names at the three hierarchical levels:
<L1 label> <L2 label> <L3 label>

L3 labels can contain one space for species and no space for genus
Eg: Order Family Species: Hymenoptera_bees Apidae Apis mellifera
Eg: Order Family Genus: Hymenoptera_bees Apidae Bombus

==================================================================
Content of ZIP files for the test dataset (insect classification):
==================================================================

classifierTestV6.zip (202MB)

MAMBO and Arthropod: NI2_MAMBO 67 taxonomic groups
Orchard: Orchard - 44 taxonomic groups
RT-Impact: RTNI - 21 taxonomic groups

Every subfolder is named, where space separates names at the three hierarchical levels:
<L1 label> <L2 label> <L3 label>

==================================================================
Content of ZIP files for the tracker evaluation dataset (insect tracker):
==================================================================

trackerTestData.zip (1.3GB)

The ZIP file can be downloaded from: https://anon.erda.au.dk/cgi-sid/ls.py?share_id=AO4fCuU3fi

The dataset contains 110 minutes of time-lapse recordings at 0.5 and 0.33 fps, comprising 1,792 frames, 2,545 detections, and 487 ground-truth insect tracks.

See the README.md file included in the ZIP archive for further details about the dataset and usages for tracker evaluation.