Automatic Photography with Google Clips | Hacker News Reader

Automatic Photography with Google Clips

ai.googleblog.com

80 pointsdsr1242 comments

Open articleView on HN

Seeing Google's approach to quantifying life's special moments in this blog post brings out the luddite in me.

When we had our first child every moment we captured felt special.

It was awe inspiring to watch our little human encounter and respond to the world.

The big ones like the first step, the first word, the first laugh.

And the smaller ones like encounters with new friends, grandparents, first time at the beach, etc.

After we had our second it was strange to see the similar reactions.

And we realized how many of these milestone responses were innate. And predictable.

It wasn't necessarily less endearing but a little less magical as parents.

Google's approach is spot on.

And it makes me sad that life's "special moments" aren't just predictable, they're inherently quantifiable by very basic AI.

As we barrel towards an AI enabled future it's going to be bittersweet to unwravel & index the human experience.

Change is always frightening.

Frequently for good reason.

I don't really get the point you're making. The AI isn't aware of parental concerns, preferences or human behaviour specific to childhood activity. It's evaluating "interestingness" visually, after someone chooses to put the camera in front of an already interesting active area such as kids playing.

The AI is not waking up and saying to itself, "initiate full power mode immediately because I detect children playing in an interesting manner".

While I love new photo gadgets, the 3 hour battery life is a little disappointing. I do like the general idea though, although what were they thinking not adding a microphone?

Incidentally, I wonder what machine learning would take from reading your post. Would it conclude that one sentence per paragraph is a good thing worth replicating? I hope not!

That’s the problem when most laypersons (including many in tech) talk about AI. It adds noise to what machine learning actually is today, and makes it hard to argue about the ethics of it when most people don’t know how to frame the problem.

I hadn't thought about it in such personal terms. Sadly, increasingly, algorithm == fate.

Google PHB: How can be bring in more data from the real world that we can analyze, categorize, and monetize?

Google dev: Photos are our most powerful data gathering tool.

Google marketing: People go lots of places and do lots of things that they don't photograph. How do we fix that?

Google PHB: We need to make it so they never ever stop taking pictures for us.

Google marketing: We'll make an app called "Clips" and tell them it's to make their lives better, while it takes pictures of all kinds of things around them that they didn't intend for anyone to see!

Google PHB: And we'll market it as "AI" because people will fall for anything if you call it "AI" or "Machine Learning."

Google management: Great idea! Raises all around! Except for the dev. Just throw another squeaky toy in his playpen. He'll think it's a bonus.

I know you meant it as a joke, but deep inside I feel there might be some truth to it.

I have one of these. It's neat, but the low frame rate for video makes the clips themselves look choppy and genrallly subpar compared to phone video. The "editorial" choices the camera makes are usually pretty great, though, erring on the side of conservative even at the higher settings.

I guess they're targeting the shareable gfycat style of clip. Silent, looping, eye-catching. I agree I'd prefer better framerate. I'd also prefer a mic. It's a shame that "privacy" may have been a factor in omitting the mic, when it should be a choice for the user.

What's the minimum focus distance? They don't say on their tech specs page.

If I understand it right, the best use-case would be a party or event where I could place the Clips camera pointed at the center/stage. The camera would record for 3 hours and then at the end I could sync it with my phone to get the best clips.

I can imagine having 3-4 of these 'Pick me up and point me at something interesting' cameras for guests at a wedding. Instead of a human having to filter through hours and hours of video, wedding photographers could offer a set of cameras for $100/hour and then use the clips to make a short video with the most interesting clips.

Yes, but it only captures short motion pictures with no audio.

As anyone who's been near a film set will tell you, capturing good quality audio in a busy environment is virtually impossible. You'll never get good sound using today's technology in a form like Clips, so that it doesn't really matter that it doesn't record it. Just dub some music or commentary over the top later.

Tell that to the billions of people already happily capturing video with sound on their phones. Considering playback is often a phone speaker, it's not like people require "film set" quality audio recordings. They just want to hear audio from the recorded location, no matter how rough. In saying that, I understand the Google Clips position with no audio, but I'd likely skip it because of this.

Well, you can always add the audio manually from the original clip. If it marks the timestamps, this could be done by a simple mp4box script.

The Google Clips camera does not have a mic, there is no audio in the original clip.

I've thought for a while that the right thing, given bulletproof privacy controls anyway, was a wearable camera that just captures 24/7 video/audio with retrospective analysis for significance.

I don't know of anyone I'd trust with such a stream, though. It would have to be on device, as they say.

but if it is only on device, (the video) then it is easily possible to get the device damaged, lost, broken, robbed in order to stop that retrospective analysis by a bad actor.

My goal is not "preventing bad things", but rather "remembering interesting things".

What if the data was synced between a small circle of your devices using end to end encryption and each device itself had full disk encryption with a strong security model to back it?

Prevents against most common data loss scenarios while protecting the information from bad actors.

Or a good actor. Let's not pretend it's only good actors that want privacy, and bad actors that want access.

i meant robbed or damaged by a bad actor, Analysis I meant for user's own use by user.

Microsoft developed such a camera years ago (sensecam) . Ahead of its time, or possibly the hardware at the time didn't make it feasible.

Edit: it was called the sensecam https://en.m.wikipedia.org/wiki/Microsoft_SenseCam

I thought about that too, but ultimately decided it wouldn't be ethical. And probably illegal in many situations.

I've been to a few events (including pro privacy events). My understanding is that all you need (in the US) is to make them aware with a sign that says by entering this room, yada yada yada.

A wearable doesn't record only in your room, though; it goes everywhere you do.

Black Mirror S01E03. Even with bullet-proof privacy, not a future I would want to live in...

Much like there is a XKCD for everything, there'll be a Black Mirror episode, too :)

What I saw in that episode was people being assholes to each other, the fact that they also happened to have a tool to prove things wasn't really the major factor. Especially the final point was a bit off to me, sure, it happens that we would be happier not knowing specific things, but being unsure about things can be pretty terrible as well and just as well destroy life.

We don't expect people to show us their diaries, we don't expect people to retell their awkward memories. We even generally try to gauge the situation before retelling a story about a friend (maybe their embarrassing blackout last weekend is not what we talk about with their boss). This is all in the social rules and conventions, and doesn't have to do with limitations in our ability to do so.

We already have the technology to record all digital communication. Seems like most people can handle this just fine.

I wonder why it works only with the Pixels, iPhones and the Galaxy 7 and 8. Is there some special hardware in those phones that any other phone doesn't have?

Likely Bluetooth 5 if I had to guess. Much higher bandwidth

Tech specs say it requires wi-fi direct and Bluetooth le. The latter is part of Bluetooth 4 spec. The tech specs also say it requires Android 7.0; nothing of specific phones (or iPhones)

Ok, well there goes that theory

From the store page [1]:

"""

Compatible Android devices: Google Pixel, Google Pixel XL, Google Pixel 2, Google Pixel 2 XL, Samsung Galaxy S7, Samsung Galaxy S8.

Compatible iOS devices: iPhone 6, iPhone 6 Plus, iPhone 6s, iPhone 6s Plus, iPhone 7, iPhone 7 Plus, iPhone 8, iPhone 8 Plus, iPhone X.

"""

And the support page [2]:

"""

Google Clips™ currently supports the following Android devices running Android 7.0 Nougat and above:

Pixel and Pixel XL

Pixel 2 and Pixel 2 XL

Samsung Galaxy S7

Samsung Galaxy S8

Google Clips currently supports the following iOS devices running iOS 11 and above:

iPhone 6 and iPhone 6 Plus

iPhone 6s and iPhone 6s Plus

iPhone 7 and iPhone 7 Plus

iPhone 8 and iPhone 8 Plus

iPhone X

"""

[1] https://store.google.com/us/product/google_clips_specs?hl=en...

[2] https://support.google.com/googleclips/answer/7545354?hl=en&...

Annoyingly the tech specs part of that page says it requires Android 7.0, then the requirements page says something different.

Given that it supports older iPhones it won't be Bluetooth 5. Android has notoriously difficult to develop for Bluetooth LE support, due to the large number of combinations of drivers, low-level Bluetooth stacks (last time I looked into this there were 4 contributed by different chip vendors and manufacturers) and API iterations. It becomes really difficult to achieve even basic tasks never mind something a little more unusual such as downloading larger amounts of data as a background task. Though it is certainly ironic to find Google struggling with this as well...

Am I the only one who thinks the advertised pictures on the site look awful? Low focal length, high aperture, no artistic look and feel at all. Why would I want such low quality looking pictures? With such images it seems to be more a device for scientific purposes than for capturing valuable moments of my life I want to frame and hang in my kitchen.

How do they compare with the typical smartphone camera picture? That's what they're aiming for here. Building an automated DSLR would raise both the price and the ML difficulty substantially.

Given your stated preferences, I am going to guess that you are not the target market for this. People who post on Facebook or order photo books from Shutterfly are.

The images advertised look worse than those of mid-range smartphone cameras. Compare those images with images of e.g. an iphone se and you will see that the focal length of google clip distorts people to a level of ridicule.

Do you think there is a market for this camera? Because I really think there is not: Extra gadget, needs to be setup anyways and, above all, poor image quality (see advertised images), not en par with mid range smartphones.

Let's wait for next year to see if that product worked.

it's so weird that they are selling a camera but doesn't list resolution in tech spec but pixel density.

also for that price, one buy 3 raspberry pi with camera achieving same function

It's not the same function. In fact, TFA is all about a feature that you don't get with a raspberry. Also, you also need a battery, and you won't fit the whole thing in the same size.

I agree their spec page is missing some specs.

Resolution appears to be 1080 x 1080 - after downloading the clip on their page. I assume this is max resolution but who knows.

One of the example videos is actually 30 fps, they must have bumped up the framerate before publishing for some reason, or maybe the app does this by default... https://mannequin.storage.googleapis.com/austin-overview/sma...

Reply on news.ycombinator.com