12.7 azface detect hats pipeline

20210406 Let’s use a pipeline to count the number of faces with their forehead occluded. If their forehead is occluded perhaps they are wearing a hat! Let’s check. We will use this photo again:

$ wget http://www.allwhitebackground.com/images/3/3818.jpg

First we count the number of faces that the model finds that the forehead is occluded. We saw in Section 12.3 that the model can identify forehead_occluded so here we run the model and pipe the result to the grep command. This keeps only those faces with forehead_occluded. Then pass the output on to the wc command (for word count), giving it the option -l (short for --lines) to count the number of lines rather than the number of words.

$ ml detect azface 3818.jpg | 
  grep forehead_occluded |
  wc -l
4

So there are 4 faces with forehead_occluded. But from the photo we might have identified just 3 people wearing caps. So let’s check who have been identified as having their forehead occluded. This gets a little more complicated since we will draw a box around the face for each of the 4 faces of interest. See Section 4.1 for an explanation of this more complex pipeline. Replace the link to the photo with any other file and you can mark up any photo.

$ ml detect azface 3818.jpg |
  grep forehead_occluded |
  cut -d, -f1 | 
  xargs printf "-draw 'polygon %s,%s %s,%s %s,%s %s,%s' " |
  awk '{print "3818.jpg -fill none -stroke red -strokewidth 5 " $0 "bb.png"}' |
  xargs convert

This produces the following file bounded.png, noting the outlier has hair covering her forehead—she is not actually wearing a cap. We always need to be careful with our assumptions!



Your donation will support ongoing availability and give you access to the PDF version of this book. Desktop Survival Guides include Data Science, GNU/Linux, and MLHub. Books available on Amazon include Data Mining with Rattle and Essentials of Data Science. Popular open source software includes rattle, wajig, and mlhub. Hosted by Togaware, a pioneer of free and open source software since 1984. Copyright © 1995-2021 Graham.Williams@togaware.com Creative Commons Attribution-ShareAlike 4.0