Our sample set of captured images is available for download here.
For this next section, we’ll use those photos to walk you through how to process and optimize models for a story. In the previous guide on Capturing Images for Photogrammetry, we took photos of Freeman Alley to create a photogrammetry model. It will also clamp the estimated texel into a range defined by minimal texel size (for high-resolution areas) and maximal texel size (for low-resolution areas). Then the Realit圜apture unwrap algorithm will use the 3D model, camera poses and input resolution to calculate optimal texel sizes.
With this style, you can define ranges where your texture must fit. The Adaptive texel size style can also be useful, especially when your scene includes varying amounts of detail, such as a statue captured in high detail, with a background that includes less detail. (Note that we’ll revisit these settings during the decimation phase.) This setting creates results that are fast and sharp, whereas the Photo-consistency-based style gets results that are slower and more complex. This will help to maintain a fixed material quality.Ĭoloring/texturing style: Visibility-based With a fixed texel size, the whole model (regardless of the resolution) will be unwrapped with a chosen texel size, and the texture of the whole scene will be rendered at the same resolution. For more detailed information, see Lens Distortion Models In Layman's Terms. However, if the processed photos are low contrast, it can be helpful to increase the sensitivity, while if the processed photos are high-contrast, a lower setting may work better.ĭistortion model: For us, Brown3 with tangential2 distortion models have been effective settings. This happens because detecting an increase in feature points means the quality of the feature points will go down. High and Ultra will detect more feature points, but may introduce issues and result in misalignment. Medium detector sensitivity generally works well. In general, we never go above 80,000 max features per image. It is counterproductive to set a high value here, because it will detect too many points. This defines the number of feature points used per image for alignment. As a general rule, we never go above 20,000 features per MPX. Setting your max features per MPX to 10,000 will prevent Realit圜apture from detecting too many features in a small area, which often results in misalignment. The more features per MPX, or per million pixels, the longer your processing time will be, as Realit圜apture will be attempting to render more components. Since this guide won’t cover every step of the Processing stage in detail, we recommend the following resources for additional support and documentation:įor alignment, these default settings are generally acceptable:
A powerful Windows computer (alternatively, you can use an AWS Computing Instance).Maya or Blender (to create camera paths).Photogrammetry software, ideally Realit圜apture or Meshroom.Photos from a photogrammetry capture (or you can download our sample set here).
What this guide covers: Tips and best practices for working with photogrammetry software to tackle the alignment, texturing and simplification of a detailed 3D model. Rather than covering every step of the process in depth, we’ve chosen to highlight the parts that we’ve found to be most important.Īlso note that while we predominantly used Realit圜apture to process and align our images, a similar workflow can be applied when using other photogrammetry software. We are making the assumption that if you’re following this guide, you already have a familiarity with 3D rendering techniques.
Note, however, that photogrammetry software can be quite complicated. All together, these steps will produce a realistic photogrammetry model.īelow we’ll lead you through each step of the process. We’ll then reconstruct the geometry of the scene to create a mesh, and finally we’ll create a detailed texture overlay to go on top of the mesh. To begin, we’ll align all of the images you shot in the capturing phase using Realit圜apture, a software we rely on throughout this guide, to create a point-cloud.
Processing photos to create a 3D scene using photogrammetry software is a multi-step operation that requires some existing knowledge, plus some extra finesse and care.