{"id":39459,"date":"2017-12-07T12:34:50","date_gmt":"2017-12-07T11:34:50","guid":{"rendered":"https:\/\/blog.buzzoole.com\/?p=39459"},"modified":"2018-05-28T11:42:11","modified_gmt":"2018-05-28T09:42:11","slug":"first-image-recognition-system-influencer-marketing","status":"publish","type":"post","link":"https:\/\/blog.buzzoole.com\/en\/buzzoole-2\/first-image-recognition-system-influencer-marketing\/","title":{"rendered":"The first image recognition system for Influencer Marketing"},"content":{"rendered":"

In our mind, <\/span>eyesight<\/b> appears to be something simple. It is easy for us to understand whether the image we are looking at is actually an animal, a person or a landscape at sunset.<\/span><\/p>\n

<\/p>\n

However, what may seem natural to us, might be an extremely complex process for machines. Yet, in recent years there has been an unprecedented development thanks to progress in the field of <\/span>machine learning<\/b> and in particular, with regard to<\/span> deep convolutional networks<\/b>, which are now able to match and – in some cases – outdo human performance.\u00a0<\/span><\/p>\n

Several studies have shown considerable progress in visual recognition in machines through validation of their work on <\/span>ImageNet<\/span><\/a>. Research after research and model after model, increasingly advanced architectures have been developed, allowing us to obtain extremely accurate image recognition. <\/span><\/p>\n

GAIIA’s eyes <\/b><\/p>\n

Since we already have a well-structured convolutional network, our task is to continue training it by focusing on the aspect that interests us the most, that is the Influencer Marketing industry. What does this mean? A “generic” image recognition system has the potential for a higher risk of error than a highly specialised one.\u00a0<\/span>What we are trying to do at Buzzoole is to develop vertical models for specific sectors, such as food, fashion, travel, etc. In this way, we will not only be able to meet the needs of the Brands that will choose to work with us, more effectively but also to pick profiles with maximum affinity.<\/span><\/p>\n

We are able to do this thanks to our very large database of Influencers and to an even greater number of images that we have processed in anonymous form in order to improve our image categorisation system. Another key aspect of our technology is that we can retrieve images as well as the <\/span>context<\/b> in which they are produced. This huge amount of information collected from various social channels, combined with a categorisation process through <\/span>NLP<\/b> (Natural Language Processing), helps us gain a deeper understanding of our users. <\/span><\/p>\n