Assembling the final pieces of the skeleton took a bit of time, mostly because of the number of bones there are and how to place all of them. There was also a setback that I didn't notice until using photographs of the Thoracosaurus display case to guide me in placing the bones. It turned out the 11th vertebrae had been skipped in scanning somehow. I had the 10th and the 12th, but not the 11th. I started to freak out a bit, but realized that about a year ago, Mark Petrovich had scanned half of the skeleton with an older, clunkier software. Luckily these scans are still backed up in multiple places. Mark and I then had to find these bones that were skipped and piece them together in PET (Polygon Editing Tool); which has a tendency to want to crash every time a user saves. I also want to note the importance of labeling and documenting the process of scanning. The students who were in charge of the scanning had a tendency to not adhere to strict naming conventions, which caused much confusion and might have been responsible for how a bone was skipped. I had also noticed some things labeled incorrectly, such as some of the skull bones. The basioccipital from the skull was also labeled as one of the dorsal vertebrae. There were also a few Caudal vertebrae that I noticed were somehow scanned out of order. I was able to sort this out though after using the reference photography. These are all reasons to make sure that whoever is in charge of a digital paleontology like this must have some sort of background and interest in the subject before diving in. This will reduce the number of errors that take place. Not just any digital artist should undertake such a daunting task. I have been trying to correct some of these mistakes by starting to go back into the original data files and renaming files and organizing them correctly. I don't want to confuse anybody who attempts to work with this data down the road after I've graduated from Drexel.
Back to the assembling front, the bones (even though severely reduced in resolution) had a tendency to be very processing intensive in the viewport of Side Effects Software's Houdini. In order to more efficiently place the bones, I set up a "switch" for each bone. This switch allowed me to place the lowest resolution version of the scan, and then move a slider to swap out the low resolution geometry for any of the other resolutions. This let me use much less computing power to place the bones initially, and then swap them for more detail for continued placement finessing as well as rendering. Then, I set up an interactive slider to control all of the switch nodes. This made it very easy to just swap all of the geometry for the entire skeleton at once; to lower or higher resolutions as I needed. This is what brought me to a point where I could go in and start modeling the rest of the bones. Since my comfort zone for modeling lies with Autodesk Maya, I had to then transfer the final pieced and reconstructed skeleton from Houdini to Maya. I simply exported an .FBX file which contained all the geometry. When importing this .FBX into Maya, the geometry came in as a singe giant piece, which would not work for adjusting placement, and rigging the creature down the road. I had to make sure to use the Mesh > Separate option in order to split the geometry back up into the individual objects. This option basically just takes any pieces of geometry that are independent from one another (i.e. the vertices aren't merged) and splits it up into separate objects based on these boundaries. The wonderful thing about once this was done, is the 10K resolution for every bone runs extremely smoothly in Maya. This is because Houdini was referencing external .OBJ files for each and every bone, where as Maya keeps this geometry recorded in the .mb file. Once this was all set up, it was time to start modeling the missing pieces.
II. Modeling the Missing Bones: Part I: The Skull
Knowing it would be the most difficult part to model, I decided to tackle the skull first. The skull is also the place with the most distinguishing features for crocodylians. The difficulty of this task was compounded by the fact that there is not much literature to be found on Thoracosaurus neocesariensis in particular. Since there is little information to be found strictly on Thoracosaurus I had to base the shape of the skull on the closest relatives I could find in the literature. The main two that I based my recreatd skull off of are Eothoracosaurus mississippiensis (Brochu 2004) and the modern Gavialis gangeticus. So this places it somewhere between a very basal gavialoid which thoracosaurus is very close to, and modern gharials. One major reason for using Eothoracosaurus is because of how in Brochu's description of the species, he is actually reinterpreting what Kenneth Carpenter described as Thoracosaurus neocesariensis in 1983. Brochu points out key differences between the osteology of this specimen and Thoracosaurus. His paper also includes a plethora of photographs of Eothoracosaurus complete with scale markers. This allowed me to import these images into Maya and set them up on image planes for me to use as cross sections to model from. I had a very good top and bottom view of the skull as well as a great top view of the jaw from the type specimen of Eothoracosaurus. Brochu's paper also included a side and back view of a braincase of another Eothoracosaurus specimen. I have noted and tried to incorporate all the relevent differences noted between the two species. For the pieces I didn't have reference images from Eothoracosaurus, I tried to use gharial skull photographs. I found some great 360 degree turntables of a gharial skull that was CT scanned. It can be found on the website for the Digital Morphology database at the University of Texas at Austin (www.digimorph.org). Most of all, even though I modeled off of this great reference imagery, I made sure to to use the contours of the actual skull scans to drive the appropriate shapes in the skull. 
The area of the skull ventral to the braincase (pterygoids, transpaltine, etc) of the skull was very difficult to model since the orthographic photographs do a great job of compressing any sort of depth. This is the one time in my life where I actually see a point to stereoscopic images, and also how they could really help paleontology (check this out). I did the best I could from a number of orthographic photographs as well as posed perspective reconstructions of crocodylians.
When creating the snout, the gharial skull was much longer than the Eothoracosaurus skull, so there was some questioning on how I should tackle this issue. I opted to go with the longer skull, but for a reason. In my research of gavialoid crocodylians, I read that snouts in modern gharial are allometric (Whitaker 2007). Apparently as gharials grow, their snouts become shorter and blunter over time. In the fossils of our Thoracosaurus, in the skull fragments there are very clear sutures that have not yet completely fused together. This lets us know that this individual was not a full grown adult. Combining this fossil data with the analogue of allometry in gharials is what led to my decision of building a longer snout than the one found in the type specimen of Eothoracosaurus described by Brochu.
The teeth took quite a bit of time to add in. I based the number of teeth on a combination of the relative space between the scanned fossil teeth that are still embedded in a piece of the dentary, the number of tooth sockets found in the Eothoracosaurus orthographic images, and how many of those fit in the extended gharial snout. The alveoli (sockets) took a lot of time to embed in the jaw. I had to use some techniques to kill off edge loops as they radiate out from the relatively dense alveoli to the rest of the snout. It was a lot of brute repeated modeling task-work.
The most puzzling aspect of the skull was definitely the lower jaw. The Eothoracosaurus jaw is much more robust and dorsal-ventrally expanded than the gharial jaw. The gharial is much more streamlined. I started with the overall top view of the jaw which worked fine, but the side view is where the discrepancy showed up. If I went strictly off of the Eothoracosaurus jaw, it would angle down pretty far away from the top of snout, which made it look like the teeth would never be able to clasp together. So I attempted to use the articular surface between the articular and quadrate from the Eothoracosaurus and combine it with a more gharial-esque jaw. If one follows this jaw rostral, there's suddenly a problem with the actual scanned part of the dentary. It has a pretty sharp angled kink in it when viewed from above, so I placed it where it matched along the length of the jaw in the top view. In the side view this makes it appear too tall for the gharial jaw and too short for the Eothoracosaurus jaw. The only way to fit it in seemed to be to have a relatively drastic kink upwards after the scanned portion as the jaw approaches the tip.
This is when I noticed an area on the inside of the scanned dentary piece. There appears to be a crack conveniently placed at the "kink" mentioned earlier. It looks like there's the possibility that this kink isn't actually osteological. This crack looks as if this kink is a result of finding the best fit for the Paleobond glue to hold the bone together for display purposes. This is something I have noticed in my fossil preparation experience. Highly fragmented bones, when glued together have a tendency to get somewhat warped around a bit. I did an experiment with fixing this "kink," which could not normally be done easily in the physical world. I set up a deforming lattice and just warped it into a position where this crack fused. when removing this kink, the scanned dentary piece fits better if it's further back on the jaw. This then seems to fit very close to the gharial jaw reference, and creates a more gradual thinning of the snout. This is something I need to consult the experts on, but I'm pretty convinced that this is how it should be. I will stop playing with the skull until I get feedback, and move on to the rest of the skeleton. This process has been extremely difficult. It is very different than my previous character modeling experience. With that it's an artistic choice when something looks good enough to move on. With this, it's much slower to try to ensure that every bit is as accurate as can be with the data that we have.
Posts
