Vexation of Photo Editing AI
As an amature photographer interested in composites and image stylization, I love the use of machine learning/artificial intelligence now being incorporated in editing software. It speeds the workflow tremendously, but some features still need a little work.
One of those is sky replacement. The software I use now does a good job of recognizing tree branches, leaves and large buildings sticking above the horizon. However, when it comes to automatically masking thin metal framework or structures like windmills extending into the middle or top area of the image, the software still needs more time in the classroom.
The image below is a good example. Both ON1 Photo RAW 2022.1 and Skylum Luminar Neo Early Access masked the trees successfully, however they badly blew proper recognition of the blades and tower of the wind turbine, requiring a significant amount of time to correct.
A second area of AI/ML problems is replacement of utility lines. Skylum Luminar Neo -- now with a limited-features Early Access version for pre-paid customers and a more-features beta for selected testers -- seems to be at the forefront in development of this feature. Using my Early Access version, I have tested the Replace Power Lines tool on a variety of pictures with mixed and often unpredictable success.
The image of the courthouse below is an example of success. The original shot had a power line running at a diagonal angle across the front of the structure. Luminar Neo Early Access removed the line with no touch-up required. Other similar images I tested did not work as well, while Early Access did an acceptable job on photos with complex power-line structures.
Screen shot of ACDSee Photo Studio 8 for Mac map feature. The software places pins on the drag-and-drop locations. |
Comments
Post a Comment