1.7 KiB
1.7 KiB
Demo 2: Comparing two images
The comparison demo outputs the predicted similarity
score of two faces by computing the squared L2 distance between
their representations.
A lower score indicates two faces are more likely of the same person.
Since the representations are on the unit hypersphere, the
scores range from 0 (the same picture) to 4.0.
The following distances between images of John Lennon and
Eric Clapton were generated with
./demos/compare.py images/examples/{lennon*,clapton*}.
| Lennon 1 | Lennon 2 | Clapton 1 | Clapton 2 |
|---|---|---|---|
 |
 |
 |
 |
The following table shows that a distance threshold of 0.8 would
distinguish these two people.
In practice, further experimentation should be done on the distance threshold.
On our LFW experiments, the mean threshold across multiple
experiments is about 0.68,
see accuracies.txt.
| Image 1 | Image 2 | Distance |
|---|---|---|
| Lennon 1 | Lennon 2 | 0.298 |
| Lennon 1 | Clapton 1 | 0.985 |
| Lennon 1 | Clapton 2 | 0.855 |
| Lennon 2 | Clapton 1 | 0.904 |
| Lennon 2 | Clapton 2 | 0.955 |
| Clapton 1 | Clapton 2 | 0.712 |