Support

Getting Started

Volume Vision requires an iPhone or iPad with a LiDAR sensor. The app also runs on Mac as “Designed for iPad.”

1. Capture — Point your device at an object and tap to capture RGB and LiDAR depth data.
2. Annotate — Draw bounding boxes or polygon masks around objects of interest.
3. Train — Configure your model settings and train directly on-device. No internet required. The compiled Metal pipeline delivers fast epoch times with zero-allocation batch feeds.
4. Export — Save trained models as CoreML for use in your own applications.

Training Modes

• Dual RGB + Depth — Fuses color and LiDAR depth channels for detection that captures structural features invisible to cameras alone.
• RGB Only — Train with standard photos when depth data is not available.

Training Tips

• Start with 20–30 annotated captures before your first training run.
• Use the Delta Depth input channel for structural features that RGB may miss.
• Experiment with learning rate schedules (constant, cosine, step decay) and dropout to control overfitting.
• The app recommends a batch size targeting roughly 25 steps per epoch — you can adjust this in training settings.
• Training runs best on iPad with longer battery life and thermal headroom.
• Monitor the thermal indicator during training — the app will automatically throttle if your device gets warm.

Known Limitations

• LiDAR capture requires a device with a LiDAR sensor (iPhone 12 Pro and later, iPad Pro 2020 and later).
• On Mac, LiDAR capture is unavailable. You can annotate and train using previously captured data.
• Training complex models with many classes may take several hours.
• Audio classification features are planned for a future release.

Contact

For bug reports, feature requests, or questions, please open an issue: GitHub Issues