A review of the open-source photogrammetry / structure-from-motion application rhiestan/Regard3D and its companion rhiestan/build-scripts β architecture, code quality, dependencies, build, security posture, and fitness for use.
A well-engineered but now-unmaintained desktop photogrammetry tool. The code itself is clean, modular and modern-for-its-era C++11, with tidy MIT licensing and sensible threading. The problem isn't the code β it's that development stopped in 2019, it carries a very heavy frozen dependency stack (OpenMVG, OpenCV 4, wxWidgets, OpenSceneGraph, Boost, Ceresβ¦), the build path is Windows-centric and fiddly, and there are no automated tests. Excellent as a learning reference or an offline OpenMVG front-end; a meaningful modernization effort to use as a base for new work.
Regard3D turns a set of overlapping photographs into a 3D model. It is a GUI front-end that orchestrates a full photogrammetry pipeline: feature detection & matching β structure-from-motion (sparse reconstruction, via OpenMVG) β densification (CMVS/PMVS, SMVS, MVE) β surface generation and texturing. The UI is built with wxWidgets; the 3D viewport uses OpenSceneGraph. It is a single-author project by Roman Hiestand, MIT-licensed, actively released from roughly 2013 through 2019.
| Aspect | Regard3D | build-scripts |
|---|---|---|
| Purpose | The application | Windows scripts to compile the 25+ third-party libs |
| Language | C++11 (73 .cpp / 100 .h / 21 .hpp / 7 .c) | Batch (.bat) + CMake |
| Build | CMake (min 3.10) | VS2015 + MSYS2 + Ninja/JOM |
| Last commit | Mar 2019 (v1.0.0) | Feb 2017 |
| Tests | none in repo | builds gtest (for deps) |
| License | MIT, headers on 92/99 files | β |
The code is cleanly layered, which is impressive for a single-author scientific app:
Regard3DMainFrame + wxFormBuilder-generated *Base classes, dialogs under src/gui/.R3DComputeMatches, R3DDensificationProcess, R3DSurfaceGenProcess, Regard3DFeatures.R3DProject persists the pipeline state (SQLite-backed camera DB).wxThreads under src/threads/, with ~90 mutex/lock sites for synchronisation; the UI stays responsive.OpenMVGHelper (the largest file, 3.2k lines) bridges to OpenMVG; src/gl/ for the OSG view.Heavy compute (OpenMVG, OpenCV) and the external densifiers are invoked out-of-process via wxExecute (6 call sites), which is a reasonable isolation choice for crash-prone native tools.
std::unique_ptr/std::shared_ptr throughout, effectively zero raw new in app code.src/licenses/.wxThread workers with mutexes keep the UI responsive during long reconstructions.build-scripts are VS2015/MSYS2 batch files (last touched 2017) that compile 25+ libraries one by one; reproducing a build today is a significant undertaking.OpenMVGHelper.cpp (3.2k lines) and Regard3DMainFrame.cpp (3.1k) concentrate a lot of logic; harder to change safely.For a desktop tool the exposure is modest, but worth stating plainly:
wxExecute with constructed paths/arguments β fine for local use; validate paths if ever driven by untrusted project files.A pragmatic, honest collection of Windows batch scripts to compile each third-party library (zlib, png, jpeg, tiff, webp, OpenSSL, Boost, Eigen, Ceres, SuiteSparse, FLANN, OpenCV, OpenMVG, OpenSceneGraph, AssImp, wxWidgets, β¦) in a prescribed order, plus a cccl MSVC/Unix compiler wrapper and a libs_versions.bat pinning versions. It does its job, but it is single-platform, manual, and dated (2017) β no Linux/macOS automation, no container, no checksum verification of downloaded sources. This is the single biggest practical barrier to adopting Regard3D today.
Use it if you want a free, offline, GUI photogrammetry tool and can obtain a working build (or the official binaries from regard3d.org), or you want a clean, readable reference for how to wrap OpenMVG in a real application.
Think twice if you need active support, a modern/maintained dependency stack, a reproducible cross-platform build, or a codebase to build a new product on without first investing in modernization (dependency upgrades, a CMake-based superbuild or vcpkg/Conan, CI, and tests).
If adopting as a base, budget for: upgrading OpenMVG/OpenCV and the image/model parsers (security + compatibility), replacing the batch build-scripts with a vcpkg/Conan + CMake superbuild, adding CI and smoke tests, and splitting the two 3k-line files.
Independent read-only evaluation generated 2026-11-16 from a clone of the repositories at their current master. No code was executed or built during this review; ratings reflect a static reading of the source, build files, and project history, and are one reviewer's judgment.