Annotated Insect Detection Images

This dataset accompanies the paper Development and evaluation of a low-cost, semi-automated camera trap for surveying bumble bee communities by Getz et al. (2026).

The dataset is provided as a single .zip archive containing trained object-detection models, training imagery, annotation files, and train/validation file lists used to develop insect detection models generated by custom Raspberry Pi-based camera traps.

The archive includes weights for two trained YOLO11s models:

Tiling-based model: trained on 640 × 640 pixel tiled images, intended for use with Slicing Aided Hyper-Inference (SAHI).

Full-image model: trained on full-frame images resized to 1280 × 1280 pixels.

The archive also contains the image and annotation datasets used to train and validate each model:

Tiled-image dataset: 66,325 .jpg tiled images and 6,795 .txt annotation files.

Full-image dataset: 6,611 original-resolution .jpg images at 2304 × 1296 pixels and 3,401 .txt annotation files.

Training and validation file lists for each model.

Annotations are provided as .txt files with filenames corresponding to the associated images. The file lists define the images assigned to each model’s training and validation sets.