The Object selection tool is designed to help select parts of the point cloud, by leveraging machine learning algorithms. These algorithms effectively extract the desired object from the point cloud by utilizing various data sources, including the point cloud itself, associated input images, and depth maps (or estimated depth when unavailable).
- This feature is available for PIX4Dmatic version 1.53 or above.
- For improved workflows: Masks - PIX4Dmatic
IN THIS ARTICLE
How to use the Object selection tool
This algorithm comprises two distinct steps:
- Image pre-processing: This processing step computes the required data to enable the object selection tool and it is a one-time requirement.
- Object selection: After completing the pre-processing step, the Object Selection tool becomes available, allowing multiple selections without the need to rerun the image pre-processing step.
To use the Object selection tool:
- Select the Image pre-processing step, which is available only if the Dense point cloud step is selected or completed.
- Activate Object selection tool: After the Image pre-processing step is completed, the Object Selection tool becomes accessible. Click on it or use the keyboard shortcut Y. The Object Selection tool supports standard modifier keys for selection tools:
- Shift key+ Left-click adds the selected points to an existing selection.
- Alt key (Option on macOS)+ Left-click removes the selected points from an existing selection.
- Click on an object within the point cloud. It's recommended to avoid selecting points near the object's borders; selecting the center is not necessary.
- The tool will then run a new machine-learning process which may take a few seconds depending on the system's hardware configuration.
Frequently Asked Questions (FAQ)
The pre-processing step is slow; is this expected?
Yes, the machine-learning computations are heavy and complex. With a modern NVIDIA GPU, it can take between 1 and 2 seconds per image. Without a discrete GPU, the process can be substantially slower. For macOS, the acceleration hardware will be used seamlessly.
The obtained selection is very sensitive to the exact location where I click, is this expected?
Yes, the machine learning algorithm can provide different results even when clicking on a similar spot. Many factors influence the final selection. If the results are not satisfactory, it is recommended to select a neighboring point and check if that improves the result.
The selection is not very precise; how can I improve it?
Precision issues may arise from lower image resolution or small object size. Improve selection quality by adding more areas of the object with the Object Selection tool or another tool (Shift key + Left-click).
The selection tool is slower on drone projects than on PIX4Dcatch ones. Is this expected?
Yes, drone imagery requires additional steps due to the lack of depth maps, leading to longer computation times.
I'm trying to select a large object, but only a part is selected. How can I fix that?
Large objects spanning multiple images can be selected by adding more areas of the object with the Shift key.
I want to select a small portion of a larger object, but the tool selects the entire object. How can I fix that?
Increase the zoom level to precisely select the component of interest of the entire object. If needed, use subtraction modification [Alt key (Option key on macOS)+ Left-click] to refine the selection.
I want to select an object scanned from multiple views, but the tool only selects a part. How can I fix that?
Use addition (Shift key + Left-click) to select parts of the object from different sides for a comprehensive selection.
Why is the selection slower if I have depth or fused point clouds besides the dense ones?
The tool will create a selection on all visible point clouds. If you are interested in only one of the point clouds, hide the others to speed up the process