Use Computer Vision to Count the Segments of or Collect Data on Other Characteristics of Trilobites

Trilobites feature an almost dizzying array of sizes, shapes, bumps, spines, and segments. Their body plans, while all following a fundamentally similar three-lobed pattern (from side to side) and three-regioned patten (from front to back), present an incredible diversity of design. Trilobite species ranged from very little (centimeter) to very large (3/4 of a meter). Some were smooth, some were bumpy or spiny, some had complex arrangements of furrows on the head or tail, some had stalked eyes, some had no eyes at all.

All of this complexity makes working out how trilobites evolved--and who's related to whom--a difficult task. At the same time, their incredible diversity--plus the fact that they were abundant, lived in all different marine environments, and as a group were around for about 250 million years--makes them a great system for asking questions about evolution, diversification, biogeography, and past ecosystems.

Representatives of current orders of trilobites

Depending on which trilobite worker you ask, there are from nine to twelve trilobite orders, plus a bunch of sub-groups (families) that are particularly difficult to place. These orders are meant to represent large groups with shared ancestry (i.e., the groups also reflect evolutionary history), and currently include: Proetida, Asaphida, Phacopida, Lichida, Ptychopariida, Harpida, Corynexochida, Redlichiida and Odontopleurida (shown above, left to right top to bottom). There is arguably another order, Agnostida (bottom right), whose diminutive species are contested as being "true" trilobites by some members of the fossil community--at least one subclade may in fact be more closely related to crustaceans.

See more at the AMNH trilobite website (https://www.amnh.org/our-research/paleontology/paleontology-faq/trilobite-website).

We are wondering if it is possible to use computer vision to collect data related to all of this crazy diversity from images of trilobites. The museum has an extensive amount of content regarding trilobites, including a great deal of imagery. Three useful datasets would be:

1. The exoskeleton is divided into three regions from the front to the back: the cephalon (head), the thorax (middle region made up of many elongate articulated segments), and the pygidium (the tail). Segments were added to the thorax during growth until they reached a point where no more segments were added. The final number of segments varied among species. HOW MANY SEGMENTS ARE THERE IN THE THORAX OF EACH SPECIMEN? Note that sometimes it is difficult to tell when the thorax ends and the pygidium begins because the pygidium is made up of fused segments. It would be useful to also know how certain the segment count for any particular image is.
2. WHICH SEGMENTS HAVE MACRO-SPINES? For example, some species have large spines extending from one of the segments, and it appears (anecdotally) that the spine is usually attached to the third segment. How frequent is this really? Are there other segments that have them and how frequent are those? Are the species that are similar in this respect closely related (e.g., in the same order)? We might be able to use this information to learn more about developmental constraints on growth in trilobites.
3. As trilobites grew, each region grew at different rates. Because of this variation, the relative lengths of different parts changed over development. We have developed models that describe this growth, and by tinkering with the input and growth rate parameters, we can make trilobites of different size and different relative proportions. Now we need to collect empirical data. WHAT ARE THE LENGTHS OF THE DIFFERENT REGIONS? You'll notice that a lot of images do not have scale bars--so unfortunate! But we do have coarse estimates of their overall length and we can use this to convert the lengths in pixels into lengths in mm. RELATIVE LENGTHS WOULD ALSO BE JUST FINE.

Some possible solutions might be:

• A solution involving a thing. What's the best way to make this thing?

• Anything else you can think of!

• AMNH Trilobite Website: A comprehensive resource containing information about the classification, time periods, localities, and other content and data related to trilobites.
• AMNH Master Gallery of Trilobites A-C
• AMNH Master Gallery of Trilobites D-I
• AMNH Master Gallery of Trilobites J-P
• AMNH Master Gallery of Trilobites Q-Z
• AMNH Trilobite images folder: This image folder contains 840+ images of trilobites with text denoting their species.

Challenge owner: Melanie Hopkins