Google Photos, Google latest attempt on fixing what is wrong with their previous photo gallery — Google+ Photos.
Get real: how much space do pictures in your phone has taken? Mine, 1.5 years, 12 GB. My phone storage, 64 GB, available for use 56 GB, and 12 GB has already taken by camera roll. Not good. Looking at some of friends phones with 16 GB flash storage, my case must be a problem for them.
To free up space, people use cloud service to back up the pictures. It is more convenient to use cloud rather than external HDD, because you know that as long as the service is up and running, your photos are always available anywhere. Thus, I tried Carousel by Dropbox, and suddenly my Dropbox storage usage jumped to 97%. Not good.
Here comes Google Photos, and it has become my gallery-saver since June 2015.
In a world where cloud storage become main competition, Google Photos provide an intriguing approach: Free, unlimited storage for those who is willing to use Google’s propietary compression method.
||Carousel by Dropbox
||iCloud Photo Library
(initial: 2 + 3GB)
(initial: 15 GB)
||*Free | 100 GB $2/month
*Unlimited only if you would use Google compression, which preserve images up to 16 MP, 1080p videos. If you would like use original resolution (no matter how low-res), you will use the Google Drive quota. Those cloud backup mentioned in the tables preserve pictures in a “private” mode rather than sharing it to the whole world.
Picture management like Flickr or Picasa now getting less and less popular, while mostly now people use Instagram and Facebook timeline to store their lifetime gallery. Things to note, however, Instagram is more social media than picture gallery, while Facebook is a social media. Facebook is well know for their notorious compression, degrading the image quality to some low resolution image, only good to see without zooming in the details. But Facebook still hold the throne for image sharing because of the user tagging technology.
Google Photos comes with Face Grouping technology. With this technology, Google will automatically grouped similar faces to one “profile”. Google does not know whose name is this face (they claim), but Google might know these face models belong to him/her. No name is ever associated to each person. Only face, and collection of photos, all anonymous.
It is like Facebook auto-tag suggestion, but this technology work exclusively in background, without the need of user labelling. Facebook has auto-tag whereas similar faces grouped as one as given the label of one of your friend. It means that Facebook uses user feedback to determine “which face” belongs to “which person (name)”. Let’s say that Facebook use both supervised learning (user manual labelling) & unsupervised learning (done in background, without user consent)
Google Photos doesn’t use user labelling for classification. Not at all. Google Photos use purely unsupervised learning to group similar faces, all the times in background. All that user can do is remove incorrectly classified pictures from the group of correctly classified. In other words: giving mark of false positive.
For example, these are some people, according to Google Photos, that lurk in my pictures.
How they are sorted, or how they are grouped, I don’t know. All I know that Google Photos use some kind of Convolutional Neural Network with 21 layers to define who and what are things inside the picture.
The grouping is not instant, per se. In my case, it takes 2 hours after about 800 pictures uploaded to Google server before the face grouping appear. Also, the first result mostly buggy, as some of my friends are considered as one person. Need a little tweaking, but 2-3 days later, my “face collection” changing, some mistakes bit by bit becoming better, adding new person, and of course, new incorrectly classified images also appear.
By Google Photos design, I can remove wrong pictures from the album, but I can’t add new pictures to established album. Damn.
Let’s see. This is one of my friends, whom from the first day until now has been incorrectly classified as two different person.
And these are two people that in the first week are considered as one person, but in reality, two different people.
Well, well. Flawless, it is not, but defect, it is not, either. In reality, Face Grouping really shine to group and show each person timeline. What amazes me is that Google Photos can correctly classified people in a very small picture.
A little, not-so-important, note: Face Grouping is disabled by default in Google Photos. The only way to enable it is by uninstalling and reinstalling Google Photos (Android: clear app data) then initialize the Google Photos account in USA VPN.
Google is not Google if not for their (in)famous search engine. Google I/O 2015 showed how Google Photos can handle search specific item inside pictures. How does it detect object? Ask the neural network, ask what’s contained in their 21 layers of neural network. I tried some queries to test whether Google Photos can recognize objects. Not every object can be returned, but here are some queries that give reasonable result:
I am satisfied with current condition, but I believe this can be improved better and better. Animal, however, need better object detection algorithm. Most of my cat pictures are considered as dogs. Want to hear one of the worst scandal? here.
Creation, by far for me is the most interesting feature. Google Photos has integrated photo assistant which has a very simple task: give you notification each time a new creation is created. Whoa, whoa, whoa, wait a bit, what is creation?
Creation is automatically generated content composed by learning and combining pictures to create things like:
The rarest creation in Google Photos is panorama. Once, I had multiple pictures of scenery from the 53rd floor in Star Wars Day museum in Roppongi, Tokyo. Here are the pictures:
Then, suddenly this picture popped up in my assistant:
Google Photos know how mine gems inside mine of rocks. What I mean with gems here are pictures that worth more than its original looks after being “automagically” post-processed by Google Photos. These effects, unfortunately, inaccessible from the dedicated Google Photos app.
This one is interesting. Google Photos’ assistant favorite is making stories, based on multiple pictures in several day in succession. Story can be considered as normal album, but with elements of storytelling. You can edit the arrangement of pictures, giving description, and make your own stories. Here is the example of my last trip to Tokyo.
A collection of videos can be automatically made into a short movies. It is easier to watch one rather than watch many videos at once. Here is a sample movies created from 36 videos and several pictures during the event of May the 4th be with You 2014.
Picture taken in succession like burst photos are taken in consideration to be made into GIF, like this:
Most instagram users collage to include many picture at once. So does Google Photos, automatically, without you picking picture-by-picture manually.
Google Photos is a cloud service that use our (Google user) images as the input for their continous research in data mining. They said our images are safe and private, while it can be said that the images are internally used for advancement of computer vision research. Google always excel in one thing: crowdsourcing (like Google Translate contribution), and this is just another chapter for them.
How useful is Google Photos? It’s recommended, especially for those who always lack of storage due to their immense size of gallery. Face grouping is somewhat a kind of novelty, but it’s novelty will wear off soon enough. I love the creation, because it makes some pictures become prettier than it should really be.
However, do note that Google is pretty powerful in automatically mining your information just from photos. If you really care about sensitive data, don’t use this service.