The photos I ran through the cat detector program in Round 1 were sample images from Adrian Rosebeck’s tutorial. Those were what I would classify as glamour shots: face-forward, well composed, well lit, and in focus. Those are ideal for the cat detector program, which performs its identification using something called a basic set of Haar features, i.e. horizontal and vertical features but not diagonal features. I’m not interested in figuring out how that works quite yet, but maybe later when I try to improve the results. What I really wanted to know is how well this program works with the average photo of animals in their environment. I decided to use only photos with animals and no people to keep things simple for now.
Results
As you saw in Round 1, when the program finds a cat, it draws a red box and labels it. For each test below, I’m showing the source image and, only if a cat is detected, the resulting photo. Let’s check it out:
Test Subject: Minnie and Romeo
Success! No cats in the source image, none detected.
Test Subject: Bandit Looking Majestic
Success! No cats in the source image, none detected.
Test Subject: Bandit Reclining
Failure! 0 cats in the source image, but 1 identified during the analysis. Not sure what went wrong here, maybe the angle of the color patch around Bandit's eye confused the program.
Test Subject: ClaraBelle
Success! Correct number of cats detected.
Test Subject: Jazmin Sleeping
Success! No cats in the source image, none detected.
Test Subject: Jazmin Dozing
Success! No cats in the source image, none detected.
Test Subject: Robin Posing in “The Dogfather”
Success! No cats in the source image, none detected.
Test Subject: Breton's Cat
Success! Correct number of cats detected. Seriously? That's the only part that looks like a cat?
Test Subject: Deer
Success! No cats in the source image, none detected.
Test Subject: Laverne the Leopard Gecko
Success! No cats in the source image, none detected.
Test Subject: Lily the Cat
Success! Correct number of cats detected.
Test Subject: Rumble the Cat
Success! Correct number of cats detected.
Test Subject: Wendy the Dog
Failure! 0 cats in the source image, but 1 identified during the analysis. Wendy is obviously not a cat. Like Bandit above, the coloring on the patch of her nose may be what tripped up the program.
Test Subject: Zada the Cat
Failure!1 cat in the source image, but no cats identified during the analysis. This is a very nice clear photo. I have no idea why the program failed.
Test Subject: Syrian Hamster
Failure! 0 cats in the source image, but 1 identified during the analysis. The hamster's training as a master of disguise finally paid off.
Test Subject: Mindy's Cat #1
Failure! 1 cat in the source image, but 2 identified during the analysis. The program confused the cat's fur for a face. I suspect that the blurry image had an adverse effect as well.
Test Subject: Mindy's Cat #2
Failure! 1 cat in the source image, but 2 identified during the analysis. Again, not consolidating overlapping images seems to be a weakness of this program.
Test Subject: Niko and the Lightsaber
Failure! 0 cats in the source image, but 1 identified during the analysis. This is not the cat you're looking for.
Test Subject: Kittie-Kittie
Success! No cats in the source image, none detected. Yes, “Kittie-Kittie” is this adorable dog’s name.
Test Subject: Chewie Chewing
Failure! 0 cats in the source image, but 1 identified during the analysis. There must be something about the details in his claws confusing the program.
Test Subject: Chewie Eating
Success! No cats in the source image, none detected.
Test Subject: Simon in Repose
Failure! 1 cat in the source image, but 2 identified during the analysis. Once again, the overlapping boxes on a blurry image seem to be causing issues.
Thanks for the Memories
A special thank you to my friends and family who graciously supplied the photos I used in this round and those coming up in the next.
Next Steps
In Round 3, I will run the cat detector against a series of real-life photos that include both pets and people.