Skip to main content
GIS-Schools GIS-Schools
Change Detection ยท GeoAI

Siamese KPConv - 3D Point Cloud Change Detection

Compare aligned point clouds from two dates to learn spatial change classes, with clear separation between survey differences and real change.

Change Detection task illustration
Family: Siamese point-cloud networkArchitecture: Paired kernel-point convolution branchesVersion: Official ISPRS 2023 implementationLicense: BSD repository license; dataset notices apply
Siamese KPConv - 3D Point Cloud Change Detection

Exact implementation and checkpoint

Official Siamese KPConv change-segmentation implementation. The repository also contains a classification variant; this entry concerns point-level change analysis. A specific downloadable pretrained checkpoint was not verified.

How it works

Paired point-cloud branches extract geometric features for comparing two observations. The task configuration defines the change labels learned from annotated pairs.

Inputs and outputs

Two co-registered clouds covering the same area, with consistent units and expected features. Output assigns configured change classes to the processed points.

Training and fine-tuning

Prepare labeled pairs using the repository's change-detection configuration. Its documented entry points include trainSiamKPConv.py and eval_SiamKPConv.py. For custom data, inspect the loader and forward scripts instead of assuming arbitrary LAS files work unchanged.

GIS and digital twin use

A research route for construction monitoring and repeat-survey review. Keep dates, acquisition settings and registration quality alongside each suggested change.

Validation on your data

Hold out complete sites and compare changes with independent observations. Report class-wise errors and examine unchanged areas for false alarms before generating update candidates.

Limitations

Registration error, occlusion, seasonality and sampling differences can resemble change. Simulated training data do not establish accuracy on a real repeat survey.

Resources and license

Official source and training instructions are supplied. The repository includes a BSD license; inspect file-specific and dataset terms. No pretrained binary is claimed or hosted; train or obtain a compatible checkpoint before inference.

Integration and validation status

This is a workflow component, not an installed iTwin plugin or automatic asset update service. Preserve coordinate reference systems, units, acquisition dates and source provenance for spatial outputs; preserve timestamps and sensor units for time-series outputs. Review results before attaching them to an authoritative asset.

GISSchools has not run training, inference or an accuracy benchmark for this entry. Covers and diagrams are conceptual illustrations. Official resources remain external.

Official references

Examples and images

Downloads and resources

External resourceOfficial repositoryProvider: Iris de Gelis and coauthors
Open resource ↗
External resourceUsage and configurationProvider: Iris de Gelis and coauthors
Open resource ↗
External resourceOfficial Source and Training WorkflowProvider: Iris de Gelis and coauthorsVersion: Official ISPRS 2023 implementationLicense: BSD repository license; dataset notices apply

Official source and training instructions are supplied. The repository includes a BSD license; inspect file-specific and dataset terms. No pretrained binary is claimed or hosted; train or obtain a compatible checkpoint before inference.

Open resource ↗
External resourceLicense and noticesProvider: Iris de Gelis and coauthors
Open resource ↗
Discussion

Comments

0

No comments yet. Start the discussion.

Join the discussion

Leave a reply

Your email address will not be published. Required fields are marked.