Making a MOBA hero
I selected a figurine of the hero Kunkka from Valve’s Defense of the Ancients Two in an experiment for 3D modeling using photogrammetry.
Making the model proved to be a challenge using Agisoft Photoscan.
Making my small object model took a lot of time, patience, and staring at progress bars attempting to decide whether I needed to pause that particular phase and leave, or whether it was going to finish in time. Taking photographs of both chunks (right side up and upside down) was simple, transferring the images over once they were taken and starting the processing in Agisoft Photoscan. From there I followed the steps and got all the way to aligning and merging the chunks, but I ran into numerous problems trying to get the two models to overlap and align properly in the merge without having one entering at an angle, or having the models facing each other and separate. Resolving the merging issue required me to experiment with all three methods of aligning and merging the chunks, playing with the settings and attempting to determine the best solution. Eventually, using camera-based alignment resolved an issue and presented the initial model, which needed to be redone due to errors and poor quality models in the initial dense clouds.
Starting from the beginning with Align Photos also meant re-doing the masks on each photograph rather than the first in each set, as the drawn masks did not carry over to the rest of the photographs. With the masks manually added I started the processing, only to encounter the same issue with poor-quality dense clouds and elements inexplicably missing from the final product. After another batch process or two I decided, to just run each phase manually rather than having it all process in one phase. Performing each process one chunk at a time did reveal one important detail: the borders from aligning photos did not include all the details, resulting in elements of each model “disappearing” during the batch process. Manually adjusting the window borders resolved the issue and allowed the modeling to continue, removing the modeling clay and base from the models.
The modelling clay and base presented a different issue in where to cut, and how that affected the model. In the initial models I cut in small pieces and left parts of the clay in because I couldn’t tell where the clay ended on the mesh when it got close to the model, and out of fear of losing parts of the model. During a workday in class I was directed to cut the base and clay out with the rectangular tool and restore the removed elements with the other model, and did that in the re-processing.
Remaking the models provided good quality models to align, but that presented new issues. The previously-aligned model had been based on markers, which initially didn’t work until I realized that some of the markers should be on the back, rather than putting them all on my figure’s face; that resulted in the, kissing models face to face in the merging process. But the new model was showing the same issues even when placing numerous markers in different areas and sides of the model, requiring a new approach to be made. Aligning based on camera positions and merging the dense clouds produced a model which was completely combined and could be used, just requiring a high-quality mesh and textures to be layered atop the new dense cloud.
Other issues arose with the base and modeling clay randomly re-appearing despite having been cut out of both models long before the merging process, requiring more processing work to re-create the dense cloud and mesh before the unneeded elements were both cut out. Gradually the model was created and fully finished, producing a model that was uploaded to SketchFab with the exception of the bottom of his shoes. That issue was rooted in the design of the model and its initial photographing, as while the hair could be fully moved and ended in points that made it easier to fully complete, while the shoes were rounded and ended at two different heights.
Producing a 3D model of sufficient quality in a small object was a long and difficult task, requiring periodic re-assessments and evaluations to resolve issues that arose from each part of the process. Resolving the issues took fully applying the lessons from class and performing each phase of the process carefully, not just assuming that one portion was completely done or that something worked as I thought it would. Re-doing the masks and testing alignments of photographs required time to troubleshoot and analyze where the photographs were taken from instead of just what was in them, working to process in a method that captured not only the models, but all the details needed to complete the modeling.
