The new tool of Facebook will let users make much more targeted image searches. For example, when you search your old photos, you will be able to look for the images where you're wearing a green shirt or back dress, or where the people in the background are dancing. It is considered as one of the big advancements in Facebook's computer vision technology and artificial intelligence.
Facebook’s New Tool Makes It Easy To Classify Photos
On Thursday, Facebook announced the update regarding its new tool in a blog post. Facebook's director, Joaquin Candela, wrote in a post that the new AI-based image tool of the firm would recognize photos at the “pixel level.” As reported by Mashable, Facebook users would be able to get results of photos based on the content of it, even when they are not tagged.
“Until recently, online search has always been a text-driven technology, even when searching through images,” quoted from Candela's post. “Whether an image was discoverable was dependent on whether it was sufficiently tagged or had the right caption — until now.” The new search tool of Facebook will also help not just the users but also the network to better identify inappropriate, spam and unreliable contents.
Pulling content from Facebook photos and videos provides an original vector on targeting correctly. Eventually, it would be nice to see a whole integrated system where a user could pull information and relate it to its searches. The new features could also help raise the ceiling on Facebook’s potential ad revenue from News Feed through pictures.
Candela also said on the post that the "image understanding" of Facebook will be applied automatic to its alternative text. This new feature developed to describe the content of photos to all the visually impaired people. "Today we're announcing that we've added a set of 12 actions, so image descriptions will now include things like 'people walking,' 'people dancing,' 'people riding horses,' 'people playing instruments,' and more."
Lumos: Facebook's Automatic Alternative Text Tool
Together with today’s new feature for image content search feature, Facebook is also updating its original Automatic Alternative Text tool. According to TechCrunch, Facebook released Lumos, the new alternative text tool last April. All Facebook users that are visually impaired could support existing text-to-speech tools in order to understand the contents of photos. The new tool could tell a user that a Facebook photo involved an object such as stage and lights.
One of the things that the new tool cannot determine is relating actions to objects seen in the pictures. But a specific team of Facebook fixed that problem already by painstakingly labeling 130,000 photos pulled from the platform itself. The firm was able to train a computer vision model to recognize and classify actions happening in photos.
On the other hand, Google open sourced its own image captioning model last fall that can easily identify objects and classify actions with accuracy over 90 percent of a photo. While Facebook is focused on making the tool learning easy for the team integrates into their projects. This simply means improving the use of the company’s general purpose FBLearner Flow compared to Google's TensorFlow.