5 Things you should know about Deep Learning

1. Deep Learning is not magic

DL (Deep Learning) covers a very large area and is very complex to understand. DL is often compared to the process of learning that occurs in the human brain. Similarly, DL is also based on the use of neural networks. Artificial neural networks actually try to imitate the brain’s activity.
A baby aged only a few months learns to identify visual content by viewing samples and from the labeling she is exposed to from the human beings around her. For example, as time passes she understands that a man usually has short hair as opposed to a woman whose hair is usually longer and whose facial features tend to be more delicate. With exposure to a growing number of examples, she learns to recognize even more refined details that differentiate between males and females.

deep learning

Eventually an adult is able to understand visual content according to the ‘training’ process he has undergone during his childhood and life. For example, if he has seen many different types of birds, he will be able to name them properly as opposed to someone who has seen a smaller sample who will be limited in his ability to classify birds.

A major problem in teaching a network using Deep Learning is providing a high quality, extensive as possible sample collection. This is actually one of the major challenges in the training of neural networks. Another difficulty lies at the basic level of deciding what the model should understand.
There are many visual examples that are similar, a few simple examples to demonstrate:

  • A Labrador is very similar to a Golden Retriever. Do we have to differentiate between them or is it enough to just identify both as hounds?
  • A drinking glass – as there are many different types of glasses, very different from each other in look and color, it will be difficult to teach the model what a glass is and we will have to split the category into wine glass, beer glass, etc.

It is very important to decide this at the beginning stages as each such decision affects what is being taught and what training set is used and this will have extensive implications on the model’s behavior and accuracy.

 

2. Good news for Deep Learning – you don’t have to start from scratch (use existing Frameworks instead!)
The DL world is evolving meteorically and there is a wide range of platforms and infrastructures used for training models. Most tools are free and the documentation is relatively comprehensive. Some central Frameworks are Caffe, Theano, Google’s TensorFlow etc. Another positive point is that both the academic world and the industry use the same tools so it’s relatively easy to use tools from the academic world.

 

3. The not so good news – GPU is better than CPU
Training a neural model for deep learning requires many hardware resources, mainly serious processing power to perform repetitive iterations in order to update the model in accordance with the samples it is working on. The mathematical calculations that take place during training are relatively simple but there is a huge number of such calculations going on at the same time.

The difference between CPU and GPU is mainly the number and strength of the processors. CPU usually has a few (a few dozen at best) strong processors as opposed to GPU that has many (thousands) of relatively weak processors. For this reason, it is better to do the parallel training on the GPU. Just to illustrate this point, training that takes only 5 days on the GPU could take months on the CPU.

The bad news is that GPUs are expensive. A GPU with thousands of processing units could cost thousands of dollars just for the graphics card. And there may also be need for dedicated servers for the graphics card, which would involve additional expenditure.

 

4. Using existing models
One of the burning questions in the field of DL is – is it possible to use a certain model to train another model? The answer is – it depends…

In certain situations a model that specializes in a specific field can be adapted to understanding a different field. The closer the fields are, the easier it is to ‘train’ a model to understand another field. For example, a model that specializes in facial recognition can serve as a basis for a model that recognizes gender. This is called ‘transfer learning’. Similarly, in the human brain, areas that are near each other understand close visual content.

The field of transfer learning is very important and interesting as it can assist in decreasing the number of samples needed train similar/close models.

 

5. The challenge: preparing the training set
The most challenging part in training a model is the collection of the large amounts of data necessary to teach the model how to understand a certain picture.
For example, if we want to teach the concept of ‘basketball’ to a model the first problem would be defining ‘basketball’, as there are many kinds of basketball – professional, children’s, women, men, street ball etc. So it is necessary to see what is it takes to understand a specific category.

Only then, we have to collect as many pictures as possible that represent the category we want to teach. This leads us to the question – which picture represents the category and which does not? The answer is not always clear.

 

Click here to learn more about Visual API

Written by Lior Cohen, PicScout VP R&D, published in the Israeli tech blog – Geektime.

How can photographers measure their popularity in the market?

‘Easy’, you might say. ‘I know how many photos I’ve sold, right? So I know how popular I am!’. Here’s a news flash – there’s a better way to measure your popularity (as a photographer, of course)!

Using Insights for Everyone, you could discover that your photos are being used a lot more than you were aware of, which will help you monetize them but more importantly, you will be able to gain the following insights:

  • Which of your photos are the most popular, e.g. do your customers favor landscape views or shoots with people in them?
  • Where are they more popular, e.g. are they more popular in France or the US? What kinds of websites use your photos more often than others?
  • How many times were they used?

These insights and others (besides giving you a great feeling about how admired you are) can help you assess your popularity as a photographer and assist you in making your next business moves; for example, what should be the subject of your next shoot, or audiences that you were previously unaware of that you can approach to promote your pictures.

And here’s an infographic just to illustrate what we are saying:

how-can-photographers-measure-their-popularity-in-the-market

How popular am I?

 

Click here to try out Insights for Everyone

Photographers just want to make a living, ask Jim Pickerell

Leading stock photo expert Jim Pickerell recently published two very articulate blog posts about the current state of stock photography. Pickerell describes the changes the field of professional photography has undergone in the past few years and the increased difficulties photographers have simply making a decent living.

According to Pickerell, it’s much harder for photographers these days because of the following difficulties plaguing the industry:

  • Over supply of photos (at least 650 million images are available for easy licensing in various databases around the world!1“) causing a serious decline in the prices of image use.
  • A huge percentage of the images in today’s stock photo collections are never used by anyone.
  • The average cost of producing an image is much higher than the price paid for its use.
  • Photographers can no longer afford to shoot whatever they like and make a decent living selling just a small percentage of their work.
  • Lack of guidance from editors or agencies as to what to shoot, for a variety of reasons.

Considering all these difficulties it’s no wonder that in his post Understanding which stock images will sell, Pickerell points out that photographers need more than ever to get better information about the images they are selling so “they don’t waste time shooting things no one is interested in buying”.  And his other recent post – Is knowing which images get the most downloads enough? goes into specifics about the same issue.

PicScout Blog - Insights for Everyone, image analytics, analytics, photographer, stock photograph

 So, where can photographers get “better information about the images” they are selling OR trying to sell?

Analytics can provide a photographer (or anyone else for that matter…) with many insights about the use of their images across the web. Using artificial intelligence, deep learning, and web crawling technology, these insights are independent of stock photo agency download numbers, subscriptions, or revenues.

How can analytics help photographers?

Insights derived from intelligent use of analytics can provide photographers with information such as where their images are used, who is using them and how, and which of their images are more popular than others. And these are just examples.

These types of insights help photographers make more informed decisions about their business – for example, about what their next shoot should be, so they don’t “waste time shooting things no one is interested in buying”.

Click here to try out Insights for Everyone

 

Photography – how to shoot based on geo data? [Infographic]

Photography – how to shoot based on Geo data?

There are so many things professional photographers worry about: from being asked to duplicate something their client saw on Pinterest or similar, through being treated like a terrorist by Police/Army /Security Guards at any public event, and up to keeping up to date with the latest gear.

Here’s one less thing to worry about: Geo data! You may wonder – how would knowing the geographical spread of my images help me?

We’ve created an infographic to help you see how geo-data can help you with your photography and enable image monetization!

photography-how-to-shoot-based-on-geo-data

Sounds interesting? Click here to try out Insights for Everyone

 

5 hacks that will help you locate potential customers

Pro photographers – it’s “Make a Wish” time!

Say a Genie showed up and announced: “I’m here to help you with your marketing efforts… what type of information would help you most?”  Our bet would be that you would probably like to find out who your potential customers are.

Because when you know who and where your customers are, you will be able to make more informed marketing decisions such as: which subjects to shoot next, how to better arrange your website, where to advertise and how to monetize your images.

a man dressed as a genie coming out of a silver tea pot

This is where PicScout Insights (a.k.a. Genie…) comes in.  Our unique learning capabilities, perfected by years of research and experience , can help you know exactly who your future customers are. And we do so by learning the actual visual content of your images, and finding them on the web

Here’s how our Genie can help you:

  1. We can tell you who is already using your images. They could be your future customers!
  2. We can communicate with your image users for you, and assist you in monetizing your images, so you can focus on your next shoot (it’s always best to leave it to the experts!)
  3. We can tell you which of your images are more popular! This insight can help you make decisions with regards to the subjects you shoot.
  4. We can help you understand how your images are used. The types of websites that use them and where on the website they are displayed could have a direct impact on the type of visual content you should produce.
  5.  We can help you upgrade your marketing efforts! The insights we provide will also help you learn where it’s best to advertise, what photos you should promote on your website etc.

 

Want to find out more? Sign up now and our Genie will start working on making your wishes come true 🙂

Analytics ? What’s in it for me?

The art of photography has come a long way since 1826 when Nicéphore Niépce used a camera to take the world’s first photograph. The world has always put an emphasis on visual content to express ideas and thoughts, rather than texts; and in recent times with the development of technology, this has become an even stronger trend. With all these goings on, it’s no wonder that marketers have realized that images do a better job at engaging their audiences.

“OK, that’s old news”, you may say. And you’re right. You take the pictures, you invest great effort in them to make sure they are up to the highest standards, and then customers buy them and use them on their website/blog/social media platform. That works great for everyone. You even know which of your pictures sell best.

But there is so much more to be gained from the ability to analyze photos that are out there on the Web. If you could examine the usage of your photos on the Web, you would gain some meaningful insights about how people engage with your visual content, and those would no doubt lead you to making more informed decisions about your work. That’s where the power of insights comes in.

 

gettyimages-464314621-analytics-whats-in-it-for-me-cropped 

Focus your photography

Artistic inclinations aside, you still have to make a living… and no one wants to waste time and money producing images that have no real demand. Photography professionals who have access to this type of analytic data, such as the locations and usage of their content, can gain significant understandings that will help them produce images that people want.

These are some of the benefits to be gained:

  • Decide what to shoot next – You may love to photograph those winter sights, but maybe your audience prefers summer sceneries? If you know what your audience prefers, you’ll be able to make more educated decisions about what to shoot next.
  • Arrange your website better – Your website is your showcase, what your customers see. Promote photos that you know work best, and you’ll generate more business!
  • Learn where to advertise – Your advertising budget is limited. This type of data can help you focus on the best sites to advertise your work.
  • Monetize your images – Analytics can help you locate potential customers who already use your images without your knowledge. These insights could enable you widen your customer base and monetize your content

“Context is the king”

To paraphrase Bill Gates’ often quoted saying mentioned above, visual content is important, but it’s not enough to know what content is the most popular. Having the relevant contextual knowledge such as: where on a website an image is shown, the type of website it features on, where in the world it’s being viewed, etc. has a direct impact on the type of visual content you should produce.
PicScout’s new Insights for Everyone can produce these insights for you, and help you focus on making informed creative choices!

Sounds interesting? Check it out!

Good Job IDC student Hadas Segal! – facial recognition model

A really exciting mentoring project has come to an end here at PicScout! Over the past few months Hadas Segal, a Computer Science major at IDC (The Interdisciplinary Center), who elected to do a hands on project in lieu of an elective course, spent a few hours every week at PicScout designing a facial recognition model to identify celebrity photos over the Internet, and learning how to detect false positives in the model to improve its results.

Throughout her work at PicScout, Hadas received ongoing guidance from PicScout algorithm researcher Dr. Leonid Brailovsky and from Lior Cohen, PicScout’s VP R&D. PicScout’s R&D team prepared a methodical plan for Hadas’ internship, including guidelines on building the model, on asking research questions, and on how to resolve these questions. In the course of her project Hadas acquired knowledge about one of the most cutting edge Artificial Intelligence topics in the industry and in theoretical research – deep learning, worked with new tools and learned to apply creative thinking.

“We are great believers in the importance of collaboration between the academic world and the hi tech industry” says Lior Cohen. “We believe that both the academic world and the industry benefit from this type of cooperation, as new opportunities for research emerge on the one hand and the students benefit from real world hands-on experience on the other hand. We are proud of Hadas’ achievement and wish her all the best in the future.”

Click here to learn more about Visual API

 

PicScout Blog - News, Community, artificial intelligence, mentoring, Deep learning, student

PicScout is recognized for role in ending global violence against children

PicScout is proud to receive recognition for our involvement in the recently launched Global Partnership to End Violence Against Children initiated by Baroness Joanna Shields, the UK Minister for Internet Safety and Security.

In recognition of PicScout’s contribution to this important cause, we have been invited to attend a reception at the United Nations in New York, hosted by the UK Ambassador to the UN and Baroness Shields on July 11, 2016 to honor the launch of the Global Partnership and Fund to End Violence Against Children, and the WePROTECT Global Alliance to End Online Child Sexual Exploitation public strategy.

The Global Partnership to End Violence Against Children is a new cross-sector partnership that unites Governments, foundations, the UN, civil society, academics, the private sector and children themselves to galvanize action and collectively reaffirm public responsibility to bring the issues of violence out of the private sphere.

PicScout was called upon to contribute the company’s experience and technology towards creation of a global solution for the prevention, discovery and resolution of online sexual child abuse, and PicScout’s visual search and deep learning technologies were utilized in the discovery of missing or abused children, and the study of their abusers to assist in tracking them down.

At PicScout we commend these efforts to fight online child sexual exploitation and are proud to have taken part in these important initiatives.

picscout-is-recognized-for-role-in-ending-global-violence-against-children 

A Photographer’s Story

The photographer behind the iconic photos of the places we all know and love speaks to us about his first camera, how it feels to have your work stolen and what PicScout means to him.

A quick glance over Rafael Macia’s accomplished body of work provides a window to the varied array of sights and places the veteran photographer has captured through his lens. Macia’s subject of choice are the iconic locations and cityscapes that the average citizen takes for granted and which hordes of eager tourists have snapped as a means to enshrine their trip to New York, Paris, Barcelona or Shanghai.

But, miles off a tourist snap, Macia’s photographs are imbued with a philosophy that could be summed up as ‘capture the most familiar places but do it better than anyone else’. The result is a collection of photographs that are as iconic as the places they represent, to the extent that when we imagine the New York skyline or Central Park in the snow we are likely imagining Macia’s version of these places without realizing it.

It’s not surprising, then, that people have flocked to Macia’s photographs even though most of them don’t know who’s behind them. But the upshot of this popularity is that it tends to go hand in hand with high rates of image theft. People will go to great lengths to use Macia’s work – whether it’s to adorn their travel websites or to use as a backdrop to an ad for a leather sofa – including stealing them.

We interviewed Macia in his home turf, New York, where we discovered that, for a long time he “didn’t know that [he] was getting ripped off” for his images, until the agency that represents his work, Science Source, started enlisting PicScout’s image protection services. Soon, he was made aware of the far-reaching trail of his images and was receiving remuneration for illegal use – “a nice surprise”.

 

4 pieces of advice for Photo Agencies

…A response to a recent blog post by Jim Pickerell

The uncertain state of the stock photography industry has joined the ranks of topics which are often brought up in conversation with some trepidation – along with the melting of the ice-caps, the Free Tibet campaign, and Prince William’s receding hairline – when speaking about these things it’s common to feel a sinking feeling, like one feels when starting a Dickens novel – you know it’s not going to be an easy read. But put on hold the foreboding commentary about nose-diving license fees and the despair narrative about homeless photographers for a moment, and the reality is far from ‘Bleak House’. As evidenced by the optimistic turnout at this year’s CEPIC, the industry is resilient, tenacious even, and, unlike Prince Will’s hair, is making a comeback.

PicScout Blog - Insights for Business, Insights for Everyone, Engineering, , deep learning, GPU, training set

No one said it was going to be easy though. Being on the block for so many years, Stock has inevitably developed some entrenched habits, which have made it vulnerable when traditional sales and production methods have been undermined in past years. According to a recent blog post by influential photographer-blogger, Jim Pickerell, Stock’s revival or downfall will be down to its flexibility and courage – to rethink everything from pricing, to how to engage new and existing audiences, to perhaps most crucially, mobilizing technology and data to their advantage. In this post, we want to respond to some of the points Pickerell makes, and offer our own advice about the changes the industry needs to make in order both to survive and thrive. The added challenge is to make these transitions whilst holding on to the fundamentals that have made it the strong and exciting industry that it is.

1.   Simplify Pricing…even more

One appeal that Pickerell makes is for agencies to simplify their pricing models. Today’s image buyers, he says, “want to be able to use any image they purchase in a variety of ways without restrictions.” The complicated pricing brackets which limit a licensed image’s usage is putting off buyers and turning them instead towards other collections such as UGC where usage rules are more liberal. Pickerell is spot on about the need to simplify, but ironically, when it comes to unveiling his ideal for ‘simplified’ pricing, he confronts us with 3 different pricing tables. In short, he doesn’t go far enough. In order to assuage the stock absconders, there needs to be a far more drastic simplification – to 2 or 3 pricing categories at most.  Image buyers of today want pricing that’s simple, clear and versatile. Complex pricing categories are just another obstacle for image buyers and an excuse for them to take their business elsewhere.

Reflection of stock market graph in window 

2.   Like it or not, data is your best friend

Pickerell’s next criticism is of the industry’s failure to exploit the potential of data to help them showcase their most popular images and producing the sort of content that people “actually want to buy”. Whilst “[i]t used to be possible to have a general idea of the…subjects needed by customers and to take a haphazard, shotgun approach to production” and still make a decent profit, “[at] today’s prices, creators cannot afford to waste time producing images that no one wants to use.” In order to do this “[i]mage creators must have much more actionable data about what is being requested and what is selling.” Meanwhile, many agencies, he points out, still don’t even have the data, let alone the ability to analyze it to grow their business. Whilst Pickerell is undeniably correct about agencies needing to get onboard the ‘data bandwagon, his definition of ‘useful’ data as primarily that which agencies can glean from their own site, is far too limiting. A more exact knowledge of what’s selling on their site is a good start, but if agencies want to get a step ahead of the steady competition, they need to cast their net wider and analyze industry-wide trends and also cross-industry trends. People who buy and use images get their inspiration from a diverse range of visual sources, ranging from cinema and TV, to advertising, to the pictures their friends post on Instagram, Facebook and snapchat. So should agencies. Our advice to agencies is to get really creative with data, and broaden their definition of what useful data beyond their own ‘local’ and industry terms.

 

3.   ‘Best’ isn’t necessarily most popular

This brings us to another fundamental issue with Pickerell’s argument, which is his definition of ‘best’ content as the top selling content on an agency’s own collection. Firstly, individuals – who bring with them their own diverse tastes and preferences – won’t automatically want to buy something because it’s popular (you wouldn’t try to sell a hot pink Chanel bag to someone who hates pink and Chanel claiming “but it’s our most popular bag!”) So, we need to redefine ‘best’ content from meaning what’s popular to being what is most relevant, meaning what’s most appropriate to a specific user under shifting conditions, which is not something a local site’s sales stats can tell you alone.

Whilst refining search terms based on actual keywords used by your site users, refining categories to smaller groups of images (100 to 500 images that “fulfill all the requirements”) and consistently revising “the order in which the image are shown in their collections” based on what’s selling, are necessary to any collection, they are also the bare minimum that agencies need to be doing to promote their content. If agencies truly want to engage their existing and potential audiences, they need to go beyond static, old-hat statistical analysis, and consider data more dynamically. This means combining local, site-based data alongside individual users’ preferences (to make auto-suggestions based on these – as is done by many e-commerce sites), but also – and this is key – data about what’s trending in the industry and beyond, as discussed earlier. As long as the data is relevant, the more multi-faceted and creative the approach to data, the richer, more personalized and ultimately, engaging, the user experience. Agencies need to start thinking in these dynamic terms and expand their definition of what’s ‘relevant’ lest they be left behind.

Silhouette of office chair installation art

4.   Find the balance

On the other hand, agencies mustn’t get carried away with data. It’s easy, once you’ve had a taste of it, to become enthralled with data to the point of enslavement. But it’s crucial to remember that data is only ever as useful as the conclusions that are drawn from it. It sounds obvious but it’s a fact that’s too easily forgotten – having relevant data is not a substitute for talented photo researchers and great content. But, if used as a support to a good collection, a creative and motivated team, and a versatile business mindset, data could be a game-changer to an agency’s success.