Collection Download

Here we provide the Ecdysis Foundation arthropod collection which includes the The Mark F. Longfellow Ecological Reference Collection (indicated by archival columns). Images are provided as publically accessible links.

Identifications are given either as "reviewed" by our Entomologists or as identified by our artifical intelligence (AI) system. The reviewed column is set to True if the image has been reviewed, or False if the identification was generated by AI. Collection date, and generalized location information is provided. If more than one image is provided for a specimen, this can be determined using the column, 'specimen_id'. We update these files periodically. The date and time the file is generated is included in the filename of the download.

You can learn more about the Ecdysis Foundation Collection and how specimens are collected and utilized at our homepage.

AI Models

Using our reviewed images, we train a Pytorch model called MetaFormer, which we use to provide identifications for all our images. The model development process is described in the following peer reviewed publication.

• Welch, K. D., Wilson, M. E., & Lundgren, J. G. (2026). Evaluation of BugBox, a software platform for AI-assisted bioinventories of arthropods. Journal of Animal Ecology, 95, 192–203. https://doi.org/10.1111/1365-2656.70178

Many of our images are orignally scans of multiple specimens using a 35mm camera. We use an Ultralytics YOLO model to segment out the arthropods and then save those as individual images before sending them to the classification model. The entire process is described on our GitHub page.

The source code for training and inference is published open access at these links

• https://github.com/EcdysisFoundation/metaformer_ecdysis
• https://github.com/EcdysisFoundation/inference-fastapi
• https://github.com/EcdysisFoundation/ultralytics

Dataset downloads

These dataset files provide download links to the images and metadata about each image

---

Reviewed Specimens 3.5 MB

This dataset includes 121,917 images representing 117,200 specimens identified by entomologists to 40 orders and 340 arthropod families. File generated Aug. 12, 2026, 6:20 p.m..

---

All Specimens 22.3 MB

This dataset includes all specimens, even those not reviewed for accuracy. There are 689,513 images of 684,744 specimens representing 40 orders and 347 arthropod families. File generated Aug. 12, 2026, 6:21 p.m..

---

Multispecimen Scans 27.3 KB

488 mutispecimen image scans and their annotation files.. File generated Aug. 12, 2026, 6:21 p.m.

Model weights downloads

These compressed files contain the model weights and training statistics. Expertise in artifical intelligence and additional software is required to use them. These downloads are large.

---

Classification Model 873.6 MB

The classification model is a Pytorch model trained from our images. (version 7) see training code here

Segmentation Model 127.5 MB

The segmentation model, yolo-segmentation (version 14) is an Ultralytics YOLO model trained from our images. see training code here