AI tools - PIX4Dmatic
The AI tools step is a one-time pre-processing step that computes the data needed by the AI-assisted features of PIX4Dmatic: the Object Selection tool and Masks
The AI tools step is a one-time pre-processing step that computes the data needed by the AI-assisted features of PIX4Dmatic: the Object Selection tool and Masks. The results are saved with the project, so the step does not need to be run again before each selection or mask.
Features that use AI tools
|
Feature |
Options needed in the AI tools step |
Minimum version |
|
Image embeddings + Depth maps |
PIX4Dmatic 1.53 |
|
|
PIX4Dmatic 1.62.0 |
||
|
Auto-mask |
Image embeddings, plus a completed Calibration step |
PIX4Dmatic 1.62.0 (input masks created in 1.67 or later) |
Requirements
- Prior steps: the AI tools step can be enabled as soon as the Calibration step is selected. At that point, only Image embeddings can be computed. The Depth maps option is enabled only when the Dense point cloud step is selected or already completed
- Input images: the step and the AI-assisted features need the project's input images.
- Hardware: GPU with at least 4 GB of VRAM is recommended, preferably more than 6 GB. On macOS, image embeddings are always computed on the CPU, even when GPU hardware acceleration is enabled in the application settings.
How to run the AI tools step
- In the Processing panel, select the AI tools step.
- Turn on the options needed for the planned workflow:
- Image embeddings: Computes image embeddings to enable object selecion and image mask tools.
- Depth maps: Compute the required data to enable the mesh, object selection tool and the mask in 3D.
- Start processing
The results are saved with the project, so selections and masks can be created without running the step again. If new images are added to the project, the step must be run again. Only the image embeddings of the new images are computed, so the processing time depends on the number of images added. Removing images normally does not require running the step again.
Once processed, any number of selections and masks can be created without running the step again.
Note: If the project was captured using PIX4Dcatch, depth maps are included with the input data, and the Depth maps toggle remains active and greyed out.
FAQ
Is slow AI tools processing expected?
Yes. With a modern NVIDIA GPU, processing takes about 1–2 seconds per image. Without a discrete GPU it can be much slower.
Does the step need to be run again for each selection or mask?
No. The results are saved with the project and reused for every selection and mask.
Related articles