ScanDot Showcases 3D Scanning on Android via ARCore Depth — Open Source

⚡ The gist in 5 seconds

• Developer faizfahmi77 has released ScanDot — an Android app that captures 3D point clouds via ARCore Depth and opens them in a built-in viewer.
• Requires a phone with ARCore Depth support; a ready-made preview APK is available in the releases, and building from source requires JDK 17 and Android SDK 36.
• A regular phone video can't be reconstructed: photogrammetry is offloaded to a computer running FFmpeg and COLMAP.

🔍 What happened

A GitHub repository called ScanDot has appeared — an experimental Android app for capturing and viewing point clouds. The code is open under MIT. At the time of publication, the project has 81 stars, 1 fork, and 5 commits. Inside there are four directories: android-app/ with the app, processing/ with the video pipeline, tests/ with tests using the Python standard library, and tools/ with utilities.

The scanner is built on ARCore: camera tracking plus depth estimation. The workflow is as follows: enter a scan name, hit Start scan, slowly circle the object in good lighting, save, and open the result from the Results section. In the viewer, the cloud can be rotated, panned, and zoomed. Scans are stored in the app's private storage — uninstall the app and they're gone. A separate full-depth mode fills in depth gaps via interpolation; the author explicitly warns that surfaces may become distorted as a result. The app does not reconstruct video from the phone. For that you need a computer: python -m processing.benchmark prepare "input.mp4" --output "run" --fps 5, then reconstruct with the path to COLMAP and the --dense flag. The output directory must not exist — otherwise the run won't start. Building the APK: cd android-app and ./gradlew assembleDebug lintDebug testDebugUnitTest, with the result in android-app/app/build/outputs/apk/debug/app-debug.apk. Install it over the old version as an update, otherwise the app's data won't be preserved.

🧠 Our take

The Depth API in ARCore has been around for years — that's general context, not a detail from the repository. But open-source projects that bring together phone-based capture, a local viewer, and an offline COLMAP pipeline under one roof are, personally, something I rarely come across. Five commits and a single author tell a different story: this isn't a product, it's a playground. Our assessment — that's exactly how it should be read.

What does this change? For a developer who needs a quick scanner prototype, the MIT license and a ready-made gradlew save days of work. For an ordinary user, it's still a toy: it's unclear what to do with a point cloud that has no textures and, according to the author, contains holes, drift, and noise. This isn't scanned furniture for a catalog — it's "raw material" that you then need to do something with. In our experience, such repositories go one of two ways: either they grow a community and tests, or they quietly stay at 81 stars.

Who wins? Those building an MVP who don't want to buy a commercial SDK. Who loses? Our assessment: no one — the project is too early-stage to displace anything. Rather, it will become a reference cited in the READMEs of other prototypes.

⚠️ What we don't know

The main thing is missing — accuracy figures. How many centimeters of error on a mug-sized object, how much the cloud drifts during long captures, how full-depth behaves on glossy and glass surfaces — none of this is in the README; there's only a caveat that "a successful run does not confirm complete or accurate geometry." CI runs the build, lint, JVM tests, and Python tests, but not the camera, not ARCore tracking, and not GPU reconstruction. The release notes, according to the author, describe compatibility and signing limitations — worth reading before installing the APK. And yes, this is one developer and one source.

Forecast: within a quarter, the project will either gain a section with accuracy measurements and camera tests in CI, or remain at the prototype level — check the README and the Issues section.

Disclaimer: not officially confirmed — there are no independent accuracy measurements; everything said about quality comes from the repository author.

Source: GitHub — faizfahmi77/ScanDot