Here is a list of detectors already trained. They can be executed in ContextCapture, Orbit Feature Extraction Pro, Orbit 3DM Manage and Extract and Reality Data Analysis Service to run Annotation jobs.Each detector was trained:
Meaning, while running on your dataset, each detector type can only be used for the same specific type of job.
The quality of the detection will depend on the similarity between your dataset and the training dataset’s description.
If using ContextCapture, we recommend you to update your version to the latest one.In case no detector fits your purpose, you are welcome to submit a help ticket from your personal portal describing your expectations.
Name
Detector Type
Description
Illustration
Links
Cracks
Photo Segmentation
Detect cracks in concrete infrastructure to enable defect inspection workflows.Dataset used: drone + handheldResolution: around 1cm/pixGeographic area: multiple
Face & License plates
Photo Object Detection
Detect faces and license plates to enable anonymization workflows.Dataset Used: mobile mapping device - PanoramasResolution: N/AGeographic area: Western Europe
Traffic signs
Detect traffic signs to enable asset inventory workflowsDataset used: Terrestrial imagery captured by mobile mapping devicesGeographic area: Multiple
Cracks Ortho
Orthophoto Segmentation
Manholes
Detect manholes to support mapping & surveying workflows- Dataset used : Drone images in urban environment- Resolution : Around 2cm/pix- Geographic Area : Eastern Europe
Terrain
Pointcloud Segmentation
Extract ground from your reality-meshDataset used: DroneResolution: Under 70cmGeographic area: multiple
RoofsA
Dataset used: vertical/aerial mapping cameraResolution: around 30cm/pixGeographic area: multiple
RoofsB
Dataset used: vertical/aerial mapping cameraResolution: around 7.5cm/pixGeographic area: Christchurch - New Zealand
Here is a list of sample datasets. They can be used to test the detectors above and the use of services like RDAS. To use one of the examples, you must replace the absolute path inside the "References" tag of the example's ContextScene.xml file with the absolute path leading to where the images were saved.
Link
Orthophoto Segmentation / Roofs
Context CaptureRDAS
Photo Object / Face and License Plates / Traffic signs
Orbit
Photo Segmentation / Cracks
Photo Segmentation / Cracks3d
Pointcloud Segmentation / Trees
Below is an extension of primary detectors’ list.
These detectors are meant to support testing of all job types.
Their training pattern is very specific and a high accuracy on personal data cannot be expected.
Coco
Detect 90 classes for everyday life objects: cars, books, chairs, etc…Dataset used: Handheld cameraResolution: Not availableGeographic area: multiple
Pascal
Detect 20 classes for everyday life elements: cars, motorbikes, persons, etc…Dataset used: Handheld cameraResolution: Not availableGeographic area: multiple
CityA
Detect 7 classes in city environment: Roofs, vegetation, poles, power lines, ground, cars, fencesDataset used: Aerial LidarResolution: 3cmGeographic area: United States
CityB
Detect 5 classes in city environment: Roofs, vegetation, bridges, power lines, groundDataset used: RGB - Aerial LidarResolution: 20cmGeographic area: Western Europe
Ghost
Detect moving elements of pointcloud capture to clan-up mapping data- Dataset used : mobile mapping pointcloud- Resolution : 5cm- Geographic Area : Western Europe
Light poles
Detect lightpoles to support mapping and asset-inventory workflows- Dataset used : mobile mapping pointcloud- Resolution : 5cm- Geographic Area : Western Europe
Rail
Detect 13 classes for usual rail assets: signals, sensors, rails, etc…Dataset used: RGB - Mobile mapping systemResolution: 3cmGeographic area: Western Europe
Trees
Detect trees for mapping or clash prediction workflows- Dataset used : Mobile mapping pointcloud- Resolution : 4cm- Geographic Area : South America