Insect Detection Images and Models from Raspberry Pi Camera Traps

Annotated image datasets and trained YOLO11s models for detecting insects, specifically bumble bees. The data accompanies a 2026 research paper and includes 66,325 tiled images and 6,611 full-frame images with corresponding annotation files. Michael Getz created this resource to support the development of low-cost, semi-automated camera traps.

Use Cases

• Fine-tuning insect detection models based on the provided YOLO11s model weights
• Benchmarking object detection performance on small targets using the tiled and full-image datasets
• Developing or evaluating Slicing Aided Hyper-Inference (SAHI) techniques for ecological surveys
• Training custom models for bumble bee community monitoring based on the annotated imagery

Strengths

• Includes two distinct, trained YOLO11s models (tiling-based and full-image) ready for use or fine-tuning
• Provides a substantial volume of training data: 66,325 tiled images and 6,611 original-resolution images
• Annotations are provided in .txt files corresponding to each image, facilitating model training
• Clear separation of training and validation sets via provided file lists

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment for large-scale training
• The dataset size is listed as 0.0 B, indicating limited scope or a metadata issue

Provenance

Source

Michael Getz, via figshare

Collection Method

Images and annotations generated by custom Raspberry Pi-based camera traps, as described in the accompanying 2026 paper.

Freshness

Last updated 2026-05-22 06:55:35; freshness should be verified