How to Use the Facebook Image Recognition Algorithm

Learn how you can tap in to Facebook’s image recognition system using niche hashtags.

Posted 3 years ago

Written by

Julia Cohen

Posted 3 years ago

Image recognition on platforms like Facebook used to be a thing of the future – reserved for theoretics in action movies and conversations about technology to come. Now, it has started to manifest itself in our daily lives – from improving automation processes to (most importantly for us at Flick) the way that social media platforms operate. What’s more, is that the market for computer-based vision is projected to reach a value of $17.4 billion by 2023. Cool, right? But why should you care?

Because, if you understand how it works – you can use it to your advantage – like say, growing your Instagram account. What we’re going to touch on in this article is how Instagram uses image recognition to reward you when you use the right hashtags that fit a specific niche. Let’s get into the nitty-gritty.

Image recognition has established itself as a key requirement for tech firms, including your old favorites Facebook (who, as we know, own Instagram). This is a big thing for businesses like this, because it makes automation a whole lot easier, and allows platform processes to be more efficient – and efficiency is key. But, consistently improving the computer visioning software they already use takes a lot of training, as systems need to be able to recognize and classify images at scale – think billions.

Current models are pretty labor-intensive – everything is manually labeled by human annotators in order to improve machine learning. Whilst this is accurate and works well, the process is starting to reach its capacity because of the sheer amount that needs to be processed.

So, what is their best next option I hear you ask? Hashtags.

How to use hashtags for Facebook Image Recognition system

Hashtags can actually help computer vision systems like Facebook’s image recognition software learn even more – through specific classification. If you add a hashtag to your post that is specific to that image, this could help tell the content of the image. This helps recognize specific subcategories or ‘niche topics’ and different elements that exist within an image. So, companies like Facebook are looking to hashtags to implement better, more consistent and more accurate image recognition systems.

Let’s say you post an old picture from a holiday. You’re in Marrakech strolling around the souks. And you add the hashtag #tbt. Now let’s say you’re a computer. You can’t see the image and the only information you have about it is #tbt. Not very specific, is it? That’s because it’s a nonvisual concept. If a hashtag is too vague, or general, or just doesn’t describe the actual visual, it’s pretty much useless for computer vision systems. Tags work as a way of data supervision, and so, irrelevant hashtags can actually confuse learning models. Facebook’s image recognition system really wants you to use relevant, topically niche hashtags on your posts because it helps them out. This is precisely why Instagram will reward you if you do use contextually relevant hashtags.

So, how can you capitalize on this? Firstly, you need to ensure you’re using hashtags that are contextually relevant.

How do I know if my hashtags are contextually relevant?

The best way to understand this is to really look at your account as a whole, and then zero in. What sort of content do you post? Let’s say it’s fashion-based. Now, look at each individual post. How can you accurately describe that post to someone who can’t physically see it? If you posted a picture of yourself in a red satin dress in Paris, you might talk about the location, the dress shape, the material, etc.

(You can also couple this up by optimizing your post’s ‘Alt Text’, but more on that later on in the article.)

Now think of those descriptors as potential hashtags: #redsatindress, #fashioninparis #alinedress.

These hashtags are all contextually relevant and specific to the photo you are posting. Another thing you want to consider is that the hashtags you pick are relevant to each other. So, instead of posting just #paris, we’ve used #fashioninparis, because it relates to the other two fashion-related hashtags.

Try looking at your next piece of content and thinking of descriptors. Once you have your descriptors or ‘potential hashtags’, you can plug them into Flick to find hashtags that are related to these, and slowly build up a list for your next post. Let’s go through an example of how we might do this:

Let’s say we are a fashion account, and this is our next post. We’ll start to think of some descriptors: ‘midi skirt’, ‘London style’, ‘minimalist outfit.’ Now, we can imagine these as hashtags ‘#midiskirt’, ‘#londonstyle’, ‘#minimalistoutfit.’ Next, we’d log in to Flick to see if these hashtags actually work for our account size and type.

You could then take a look at which hashtags suit your post the best, and copy them to your clipboard. If you’re not really sure how to pick the right ones, take a look at our strategy guide, which will walk you through exactly what you need to do, if you want to find the most suitable hashtags for your account. You can also expand hashtags from your initial search and find more, eventually picking your top 30 hashtags that are contextually relevant to your post – voila!

We’re going to let you in on a little secret that will save time when it comes to your future posts. Once you’ve done your research and picked a set of hashtags you like, you can add them to a collection to save for later – so that you don’t have to do the research phase all over again. Create a series of hashtags collections on Flick, so that over time you can build up a library of effective groups, that can be applied to your future posts. For example, you might create a location-based fashion collection, a collection for when you specifically post pastel outfits, or a collection that would be great when describing a killer monochrome look.

You may remember that we mentioned optimizing the Alt text section on your Instagram post earlier. This is another way to communicate to Instagram what the content of your image is. Using descriptors in your Alt text section will help you broaden your audience but also work like SEO would for a blog post or webpage. The Facebook Image Recognition tools favor the use of Alt Text. For example, if your photo featured a brown furry coat, you might put this descriptor in your Alt text section. To learn more about how Alt text can become your new best friend on Instagram, check out this recent blog post we wrote, where we go a little more in-depth on how to optimize the Alt text section of your Instagram posts.