Photogrammetry - Komainu Statue
This project used 670 images from a sample dataset provided by Reality Capture. Images were taken by NRHK and licensed by the Open Heritage 3D project which is digitizing historic works across Japan. As a test for Unreal Engine 5, the model was only simplified to 1 million polygons and a 16K texture before importing with Nanite.
Built From JPGs
670 images were selected out of a pool of over 2000 photos of the building’s front facade and two statues. Images were imported into Reality Capture in batches of about 50 JPGs. This allowed me to confirm early-on that images were aligning correctly before trying to calculate using the entire dataset.
*In hindsight I should have started construction with only a dozen images, and set ground plane and dimensions with these images only. Trying to place control points while sifting through over a hundred images was much slower and less accurate.
Image Alignment
Once the initial batch of about 50 images were aligned, additional sets of images were added to refine details and fill in gaps in the point cloud.
Once a reasonable density was reached, the first poly model could be calculated. This initial statue was over 2.4 million triangles. I then reduced the model slightly and used Reality Capture’s tools to generate textures and color correct for highlights and shadows.
Import to UE5
The reduced model still had 1 million triangles and a 16K texture, and i was interested in how Nanite would handle this. Resource usage was (of course) excessive, but UE5 did not have any major issues working with the excessive FBX file.
Environment is made from the UE Marketplace City Environment Megapack vol 02. I really like the multicolored Cyberpunk-inspired lighting and was quite happy with the appearance of the texture produced by Reality Capture.
My next step will be to spend a little time in Substance Painter to enhance the textures and emphasize material differences between the statue and pedestal.
