Face Tagging - Refining the Training

You should find that MediagraphAI does a very good job of finding and tagging faces. There may be some instances where a face is not recognized (false negatives) or where some other person is wrongly included (false positives. Here are some ways to resolve these situations. 

Fixing false negatives

If there are photos of a person which are not being recognized, you can add additional training faces, up to 100 photos per person. This is particularly useful when the person appears in the photo collection at various ages, or is shown at different angles. You can use the unrecognized faces to help Mediagraph to better recognize this person. 

A workflow for adding training faces

We have provided tools to help you improve the training. You can easily find files where the person was NOT recognized. These will be a good source for additional training.

Pro Tip: If you have previously keyworded photos this person, and that keyword is in your Tag Tree, you should use these to do the additional training. 

1. Find a group of photos that include the person in question.
   1.  You may want to make a Lightbox from these faces to help in workflow.
2. Make sure you have run Search Faces on all the files. 
3. In the Filter Panel>Faces, mouse over the file counter and click on Filter>Does not have tag. This will show you all files in the set where the face has not been recognized. {screenshot}
   
   Mouse-over a Face Tag an you will get the option to show all files that don't have the tag. Clicking on the circle next to Does not have tag will show the 15 files where Josie does not appear, or is not recognized. 
4. Choose one of these files and open in the Asset Detail View. Find the face and add the name. 
5. Make note of how many files have this person’s Face Tag. 
   1. Select all the files in the group and run Search Faces again. 
   2. Refresh the Filter Panel. 
   3. See how the number changes.
6.  Repeat the process until you get the desired level of accuracy. 

Fixing False Positives

Typically, false positives are caused by the use of low-quality images for the training. You can add better views of this person’s face and remove poor quality faces. When you run Face Search again, you should see some of the false positives corrected. 

Survey the problem

There are two parts to fixing the Training Faces: removing/replacing bad Training Faces and fixing the False Positives. But first, let’s check and see if we want to do a total reset for this person. 

1. Go to Manage>Faces and search on the person in question. Click on Taggings to see all files with this Face Tag. 
   
   
   
2. If there are one or two false positives, you can try to fix the training faces and rerun the Face Search.
3. If there are a lot of false positives, it may be best to reset this Face and start again. 

Remove poor-quality Training Faces

Poor quality faces can be removed in the Manage>Faces panel. Here are the steps.

1. Search on the name if needed.
2. Click on the Trained Faces to see all faces. 
   
   Click "Trained Faces to see and manage the training for this person. 
   
3. Click the X near the face for any poor quality images. 
   
   You can remove any training faces which are poor quality, such as the third one on the top row. 
4. It’s okay to remove all training faces if all of them are poor. 

Fixing False Positives

Once you have removed poor quality Training Faces, let’s see how it improves the tagging accuracy. 

1. Find all files currently tagged with this face. You can do this from the Faces panel by clicking on Taggings. 
2. Add all tagged faces for this person to a Lightbox to help track the improvement. 
3. Select all these files and run Search Faces again. 
4. See if the face tag has been removed from all False Positives. 
5. If there are a few remaining false positives, you can remove these individually in the Asset Detail view. 
   
   In the Asset Detail view, you can remove falsely-identified faces by clicking the trash can icon. 
   
   Removing the Face Tag resets the face. 
   
   In a case like the one above, you may want to ignore the face since it's tough for anyone to tell who this is. Click the Ignore Face button. 
6. If you still see a lot of false positives - and especially if there are no better faces to train from - we suggest resetting the Face Tag. 

Resetting this Face Tag

You reset the tag in Manage>Faces. Use this to remove the tag from every file. It will remove the Person Tag as well if it’s been applied only through AI Face Tagging. 

1. Go to Manage>Faces
2. Find the Face Tag you want to reset.
3. Click the Reset button.
4. If you still want to tag this person, find a better Training Face. 
5. If there is no better training face available, you may just want to use the Person Tag and forgo the Face Tag.