Implement image area as alternative to distance to ground
Closed this issue · 2 comments
Laser points to determine the image area are a common practice for underwater images. We could extend UnKnoT to use the image area as alternative to the distance of the camera to the sea floor if the distance information is not available. The mean image area of the target volume could be used to determine the scaling factor for each source image.
This can be added as a fourth option "Knowledge transfer (using image area)" as alternative to "Knowledge transfer (using distance to ground)". In the code this could be the type area_knowledge_transfer, whereas the distance to ground type stays knowledge_transfer for backwards compatibility (also update InstanceSegmentationRequest for this). Then only a new PrepareAreaKnowledgeTransfer job is needed that computes the kt_scale_factor differently. The rest are UI and manual updates.
This can be extended even more:
- If not all source images have the image area information, offer an option to "take only annotations from images with image area".
- If not all target images have the image area information, offer an option to "include only images with image area" in the MAIA job. This depends on #49.