Tag Archives: facial recognition

Bits and Pieces

Posted on by

Some quickies for a slow Thursday.

First, a prediction I got right. In talking about Google’s addition of automatic tagging to their Photo app, I said “If the recognition works well, the advantages are obvious. If it doesn’t work well, then we’ve got a repeat of Flickr’s recent image tagging fiasco.”

Earlier this week, Ars Technica reported that the app was tagging photos of two black people as “gorillas”.

Google handled it well: they immediately removed the tags, apologized publicly, and worked with the man who reported the problem to tweak the facial recognition code.

But honestly, this can’t be the only offensive incorrect recognition lurking in the code. New prediction: we’ll see more such stories about Google, Flickr, and any other photo storage and display software that assigns tags automatically.

You may have heard that a new debate has been sweeping the Internet lately. More polarizing than what color the dress is, more riveting than escaped llamas, it’s The Great Peacamole debate!

A couple of years ago, Melissa Clark, a New York Times columnist wrote about a guacamole recipe based on green peas. The world ignored it. Yesterday she wrote about it again, and the Internet–Twitter in particular–exploded.

Tweets from both sides of the political divide condemned the recipe:

And yet Ms. Clark remains defiant:

The thing is, this recipe not only includes peas, but also, God help us, sunflower seeds.

I’m sure the recipe is as delicious as Ms. Clark claims–but it isn’t guacamole. If it had been billed as what it is, Avocado/pea dip, we would have avoided this whole debate.

But still, there’s a bright spot in the debacle. We’ve found an issue that unites President Obama and Texas Republicans. Maybe, just maybe, they can build on that agreement. Wouldn’t it be wonderful if something so wrong as peacamole led to an agreement on gun control, immigration, or abortion rights?

In sadder news, Tama, the feline stationmaster of Japan’s Kishigawa railway line, died last week. Her funeral was attended by 3,000 mourners.

I’ve written several times about cats working to promote their own selfish agendas or achieve world domination. It’s a pleasant change to take note of a cat working to improve her life by helping the humans around her.

Tama rose from poverty–a former stray–and single-pawedly saved the rail line from bankruptcy, and drew more than a billion yen in tourist income the the region. In recognition of her efforts, she’s been appointed to the post of “honourable eternal stationmaster” and has been deified.

Her apprentice, Nitama, has taken on the role of honorary stationmaster.

And finally, CNET and other venues are reporting that Amazon will be changing the way it weights reviews. Instead of simply averaging all reviews’ ratings, they’ll begin giving more weight to “useful” reviews.

Although the expect the weightings to change over time, currently the plan is to give more weight to verified Amazon buyers’ reviews, newer reviews, and reviews customers flag as helpful.

I have mixed feelings about the change. I can see it making a lot of sense in some areas. Giving more weight to newer reviews and “helpful” reviews of appliances, toys, and tech gadgets makes sense to me. As similar products come out, reviews that compare multiple options and weigh the tradeoffs should get more weight.

On the other hand, I don’t think that’s as true in other fields. Is a recent review of Twilight automatically more useful than one that was written when the book came out? Should a review of Jurassic Park that compares it with Jurassic World be granted more weight than a review from last year? How much weight does a multiply-helpful-flagged review of Madonna’s Like a Virgin from 1984 get compared to a review from 2014?

I’ll be watching to see how this develops.

Mew Are You?

Posted on by

Apparently yesterday was Facial Recognition for Pets Day. I must have missed the announcement, and for some odd reason, it doesn’t seem to be a Federal holiday, as I did have some mail delivered.

When I checked my e-mail, I found a note from Lior pointing me to an article about “Bistro”. The article in turn pointed me to the Indiegogo campaign.

Bistro is, as the masterminds behind it say, The Smart Feeder That Recognizes Your Cat’s Face. It can recognize your cats, dispense the amount of food you want them to have, display webcam footage of your cats eating, and generally do everything you would do except pat the cats.

Sounds like quite the spiffy gadget, doesn’t it? Let’s take a closer look, because quite frankly, I wouldn’t let this thing in the door.

According to the campaign page, Bistro can use advanced cat-facial recognition technology to determine which of your cats is eating or drinking, record each cat’s individual diet history (how much food and water it eats or drinks each time it visits the feeder), notify you of any change in the cats’ health status (eating more or less than usual), display the video stream from the facial recognition camera on smartphones, and “share your kitty’s life with Bistro’s online cat community”.

It works its magic by weighing the food and water it dispenses, so it knows how much the cats consume; weighing the cats every time they eat or drink; and sending all of the information to Bistro’s servers for analysis.

Why is the analysis done on Bistro’s servers? Why is the video stored and displayed from Bistro’s servers? Here’s a hint: “All features on the Bistro App will be FREE for every Bistro owner.” Catching on? Here’s another hint: “Get feeding and health advice…See cat food reviews and ratings”. Bistro isn’t saying it in so many words, but it seems clear to me that they expect to sell usage and user information to pet food manufacturers and probably other pet-oriented vendors. ‘Nuki is uppity enough as it is. Can you imagine what it would be like if he started getting mail?

Dear Mr. Watanuki,

Why do you continue to let your humans feed you that awful meat-based food that’s stale by evening?

It’s time for you to stand up for yourself and demand [name redacted] brand cat food, made with only the finest grains and loaded with preservatives to keep it fresh in your Bistro feeder all day!

Sincerely,

Corporate Drone #3752 (who is willfully unaware that cats are obligate carnivores whose digestive systems are tuned to eat meat…)

Actually cat food ads are the least of it. If they sell the usage data, their customers are going to start getting advertisements from vendors of miracle cures for renal failure.

Let’s get real here. If a smartphone can use facial recognition to unlock the screen, Bistro could easily hold enough brains to do the feline facial recognition locally and transfer the consumption and weight data to phones via Bluetooth just like exercise gear does for humans. The video could be held on the feeder on a cheap SD card (or a USB-connected hard drive for the real data fiends) and transferred over Wi-Fi. This is a $250 device folks, and it already has Wi-Fi capability. Adding BT, SD, and USB can’t add more than $15 to the cost, and the creation of the analysis software is pretty close to the same whether it’s being written and tested for Bistro’s servers or for individual users’ smartphones.

I suspect Bistro’s counterargument to doing all the processing locally is that by doing it on their central servers, they can detect changes in the data immediately and send alerts without having to wait for data to be dumped to the phone for processing. If so, I disagree: if they alert on a single feeding being smaller or larger than usual, they’re going to be generating a lot of false warnings*. Alerts should be based on series of atypical feedings, not a single, possibly-spurious event. If you’re looking for a series of events, daily uploads should be more than enough.

* Say, maybe they’re also getting paid by vets anticipating a surge in office visits. “Fluffy didn’t eat as much as usual in his midday feeding! What’s wrong with him?!” Nah, probably not. Who wants to field that many panicked phone calls?

On the other hand, I’m not a vet. Maybe there is some benefit in instant alerts. Fine. Build the thing around a Raspberry Pi instead of a custom processing board and do all the processing on the feeder instead of Bistro’s servers. Your total hardware and development costs probably go down instead of up, the alerts can be sent directly from the feeder (eliminating one possible point of failure in the process), and you can still do all of the same social functions. (Why the heck do they think anybody would want to watch videos of someone else’s cats eat?)

And I haven’t even touched on the security concerns of having all of that data on central servers. I’m not going to bother spelling it out here. It’s no different than Dropcam storing a continuous video record of your home; Apple, Google, and Amazon storing months of your location data; or Target storing your credit card information.

As of Wednesday evening, Bistro’s Indiegogo campaign was 60% funded with a month remaining. It seems likely they’ll succeed. I don’t wish them ill, but I hope I’m wrong about the demand for their product, because I doubt I’m wrong about their business plan.

OK, one product doesn’t make a national event. So why am I calling yesterday Facial Recognition for Pets Day?

After I finished digesting Bistro’s offering (sorry), I went downstairs to read the paper and I found an article about Finding Rover.

Finding Rover is a smartphone app that uses canine facial recognition to help locate lost dogs. (The developer, John Polimeno, is planning to expand to cats later this year–no word on whether it’ll be as part of the same service, or if he’ll be spinning off Finding Fluffy as a separate venture.)

You take a picture of your dog and upload it to their server. Users of the app take pictures of possibly-lost dogs which are matched against the database. If you’ve marked your dog as lost and a match is found, you get an alert.

Animal shelters are apparently very favorably impressed with Finding Rover. It’s free (advertising supported, but with ads on the website, not targeted ads sent directly to users), and users can enroll their pets without a visit to the vet to get a chip implanted.

Unlike Bistro’s case, for Finding Rover, uploading the photos to a central server makes sense: far better to do searches for matches in one place than redundantly on phones around the world. And, if properly handled, the amount of personally identifiable information can be much lower. At minimum, it could be no more than an e-mail address.

I’m going to wish Mr. Polimeno well. Our gang is chipped, and they’re indoors-only, but when Finding Rover expands to felines, I’ll give them a close look. At first glance, I don’t see a downside to having the fuzzies enrolled, just in case they get loose.

Here’s looking at you.