Open markcrowley1 opened 1 year ago
Human Aided point identification in seperate images.
While I would like to implement a system that provides automatic point detection and mapping between images, a feature could be added so that after an algorithm attempts this, the user has the option to view the algorithms attempt and correct it by hand if necessary. An example of an app with a similar feature to this is camscanner.
This could increase the practical performance of the final application.
Approximate unknown 3D information
If there is not enough information in the 2D images provided, there are algorithms that can estimate what unmapped object surfaces will look like and fill in the details themselves. This is accurate enough to provide a reasonable 3D color representation of the object from just a single 2D image.
Example of intended feature (high quality implementation): link
Category specific mesh reconstruction (CMR) could be an important field to research.
Object Detection
Some level of object detection will be necessary if a mesh reconstruction algorithm is to be implemented.
Object Classification
To implement category based specific mesh reconstruction, the category must be known - for the technique to work on more than one type of category there will need to be a process where the object is classified within a known category.
Using this Issue to keep track of ideas for applications/features.