*This feature is coming soon! Keep reading to learn more ...
Scenario
As a media manager, you are in charge of using AI engines to correctly recognize a set of people in your video and image archives. You need to create a training set of crops in order to do this. This process is cumbersome as you have to first find images from your existing archives or Internet (for celebrities such as famous soccer players) and then have to crop them based on the quality guidelines specified by the AI engine vendors.
Once you have cropped the faces, you have to upload them into the AI engines' training library. Mistakes happen. Mislabeling the cropped image with the wrong person’s name, or using cropped images have faces that are blurry, partially visible is common place. This can waste a lot of time & AI engine usage costs, as you need to update the training set again and then re-run the AI recognition engines again on your media.
Enter Crop.photo's automatic quality control (QC)
From the summary page when you click Select for AI-Training Crop.photo will automatically apply a number of QC rules to weed out face crops that would lead to poor training of your AI engines, which in the end could cost your organization an arm and a leg to redo the training and re-run the AI engines on your media set.
What QC rules does Crop.photo apply?
This list below is constantly evolving, but this is what we do automatically for you to weed out bad crops for AI training:
Remove crops do not have faces in them (animals, object landscape etc)
Remove crops with more than 1 face per crop
Remove crops that are smaller than 50x50 pixels
Remove crops that are not sharp or are blurry
Remove crops that are too dark
Remove crops where the face is titled so much that AI engines will get confused
Remove crops where the face is covered with objects such as hats, glasses, headbands, and masks