Skip to content

Trilobite Bits

Jump to bottom
Mike Caprio edited this page Mar 23, 2018 · 14 revisions

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

Hackathon Findings

Hackathon Projects

Background

Trilobites are an extinct group of arthropods (whose living members include insects, crustaceans, and scorpions). Although their basic body plan follows a fundamentally similar three-lobed pattern (from side to side) and three-regioned patten (from front to back), they are in fact incredibly diverse in terms of the size and shape of those regions. 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.

The trilobite body is divided up into the three regions: cephalon (blue), thorax (white), pygidium (pink)

See more at the AMNH trilobite website.

Solutions

We are wondering if it is possible to use computer vision and image analysis 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. HOW MANY SEGMENTS ARE THERE IN THE THORAX OF EACH SPECIMEN? 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. 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. WHAT ARE THE LENGTHS OF THE DIFFERENT REGIONS? 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. You'll notice that a lot of images do not have scale bars--unfortunate but not fatal! 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 trainable computer vision system. A system that could make guesses at which order a trilobite belongs to, or that can collect any of the previously mentioned "useful datasets" above. A human should be able to view the systems results and provide correction and training input to make the system better.

  • A database of existing characteristics. Many of the online images and resources available on trilobites already have some of this data--or at least taxonomic and locality information--and it would be great to aggregate it into a central database. Preserving the sources of any data you scrape and collect is critical for citation purposes, so please be sure to build in the ability to preserve the identity (institution, investigator names) and location (URL and/or query against a database) of the original data source.

  • A crowdsourcing system that makes it easier to collect observations from images. If the utilization of computer vision isn't practical or implementable, a crowdsourcing system that allows individuals to accurately create data and metadata about trilobites would be a viable option.

  • Anything else you can think of!

Resources

Be sure to check the Online Resources and Data Sets page to see if there might be any general purpose code or utilities you might use. What other computer vision resources are you familiar with that you could utilize? We'd love to learn about them!

SPECIAL NOTE ABOUT IMAGERY RIGHTS: All trilobite image content is copyrighted by several parties and is not to be modified or redistributed, and all local copies on your systems must be removed after the hackathon is over. This challenge is only to create data and metadata about the images, not to be used for any other purpose.

Challenge owner: Melanie Hopkins

Challenges --|-- Online Resources And Data Sets --|-- Code of Conduct --|-- Home

Challenges

Online Resources And Data Sets

Museum Stakeholders

Home

Clone this wiki locally