Adobe Catches Blind Spots
Adobe wants to help language models understand art.
Sign up to uncover the latest in emerging technology.
Art doesn’t come easy to everyone. Adobe wants to help language-based AI paint a better picture.
The company filed a patent application for what it calls a “visually guided” machine learning language model. Adobe’s system aims to help language-based machine learning models overcome the “limited visual intuition” they have trying to visibly comprehend what’s represented in text.
Essentially, a machine learning model is trained to understand and visually interpret meaning from text. The model is trained with digital images and text associated with those images (i.e., photos and their captions). The model is trained in a way that causes “similar visual concepts to be clustered together,” allowing it to identify items that are in similar categories but may be described in different contexts.
Once the model is trained, it can be used for a variety of functions, including text classification, natural language understanding, digital content searches and text summarization. While this sounds technical, think of it like this: If you give Adobe’s trained model an image of the beach scattered in chairs and sand toys, then ask it to remove “beach accessories,” it’ll identify and remove those objects.
Adobe said its system bypasses a common issue that conventional language models face: when digital images are visually similar, but the text describing them is different. The company noted “systems that rely on conventional language models to support visual concepts may encounter inaccuracies and fail in some instances.”
Adobe has been working hard at making AI a focus of its suite of audio-visual tools. As the company noted in its filing, this new tech could support a host of AI-enabled photo and video editing. But securing patents on any visual-based AI tools could give the company a competitive advantage, both in the creative tools market and the general battle for AI dominance.
The tech firm seemingly sees the benefit of scooping up visual AI patents. Most of Adobe’s latest published patent filings are for AI-based tools, including tech that improves machine learning for prediction and document rendering, and a method for efficiently training neural networks to better catch blurry or less-obvious features in photos.
Outside of patent filings, the company has already released several AI-enabled features. Most recently, the company launched its own AI image generator called Firefly on a few of its apps, which it claims is “designed to be safe for commercial use” as it is only trained on content the company is licensed to use. One of its first Firefly-backed tools is Generative Fill, which can expand an image’s borders or add in new objects and is expected to be available to all photoshop users later this year.
While the company’s Creative Cloud suite is widely seen as an industry standard, Adobe still has incentive to keep up with the rising tide of AI and stay on top. Canva, one of the tech firm’s only competitors, recently debuted its own suite of AI-powered tools, including generative design and copywriting assistants.
Plus, while Adobe doesn’t directly compete with top players like Google and OpenAI, AI is occupying the brainspace of anyone that has a stake in the world of tech. Every tech firm worth its salt seems to be grasping for some way to implement AI, whether they need it or just fear being left behind.
Have any comments, tips or suggestions? Drop us a line! Email at admin@patentdrop.xyz or shoot us a DM on Twitter @patentdrop. If you want to get Patent Drop in your inbox, click here to subscribe.