A rising challenge for marketers is awareness of, and keeping up with, the enormous volume of videos that may contain highly relevant content, or even their own branded content. Moreover, marketers increasingly need scalable alerts on potentially unauthorized, but possibly highly collaborative, video content circulating.
For example, consider an advertiser that wants to target any content associated with Spiderman. The advertiser could be Marvel themselves, whose marketers need to know if unbranded short-form videos are trending on the internet so their team can quickly take action, whether proactive or defensive. The advertiser could also be a partner toy manufacturer that is interested in positioning its advertising against any brand-safe Spiderman content.
But the deluge of branded, and even more commonly organic, unbranded, video is unmanageable solely through human review. Without the right tools, no team can effectively tackle the sheer volume of video circulating across the globe.
What is Active Learning?
Active Learning is a form of Artificial Intelligence (AI) that allows human users to quickly relay specific datapoints or image attributes of information into AI-based algorithms with a desire to identify patterns and output similar examples so that the CV can identify that object or event in a corpus of content.
For example, in the Spiderman case, the human user can provide 10 or so samples of Spiderman images into the system. The AI-algorithms identify the closest matches, which the human user can review to provide feedback. The AI-based algorithm receives the feedback to continue to find the targeted event more precisely. Additional iterations can occur between the human user and the AI-algorithm. Thus, the algorithm actively learns and continues to incorporate human feedback to create closer and closer matches to the right identifications.
Advertising is not the only applicable use case. For example, in the emerging cannabis industry, a cannabis farm may need to monitor a high volume of plants. The risk of losing a crop is a concern, and a crop's output, performance, and profit are not guaranteed. With AI and Active Learning, simply by adding video to monitor crops can provide much more informed monitoring, intervention, and cultivation. With cannabis, Active Learning simply needs images of “healthy,” “unhealthy,” “dehydrated,” “overwatered,” “mold” examples to start the learning process. High-risk alerts around “bugs” can also be trained into Active Learning. Over time, examples of plants that emerged as high-yielding or high-quality can be incorporated to continue training the algorithm to predict and intervene when necessary.
The most critical advantage of Active Learning is that the AI-algorithm can exponentially scale the speed of learning and adaptation. The AI tool becomes increasingly selective and precise at machine-level speeds versus image-by-image human review.
How Active Learning and Computer Vision (CV) enables marketers to tackle the constant stream of video content
The crux of a solution is to create “segments.” With Netra’s Active Learning and CV, segments critical for marketers can be created through AI. For example, for Spiderman segments, AI can be quickly trained to identify Spiderman-like elements and then refined: Spiderman’s uniform, “webs,” scaling buildings, Marvel logo, etc…Once these elements are identified and trained through Active Learning, Netra’s AI creates a Spiderman “segment.” As short-form video content is scanned through our API with CV, any video that contains a Spiderman segment through our frame-by-frame analysis is identified and tagged. Other elements around scene and context are also identified and relayed back through the API.
What challenges do such custom segments solve?
The most immediate solution for marketers is the ability to identify the appearance of a targeted segment, whether identified in second-:01 or second-:45 of a video. No other solution offers the ability to create segments that are identified within the imagery contained within a video’s frames.
From this segment identification, marketers can quickly respond to the opportunity identified within the video, whether to proactively place a highly targeted ad or to flag the video to the marketing team. In Marvel’s situation, when a Spiderman segment is identified within a video, the marketing team may choose to simply ignore the video, it may decide to place its own advertising against it, or as the case may be, it may be highlighted as a high brand-safety concern.
For applications outside of marketing, segmentation can be applied similarly. In the cannabis example, potential segments could include “high” and “low” quality plants, with an aim to drive all plants towards “high” quality and address the plants that are emerging as “low” quality. This information on where plants are currently classified can be recommended to the grower to inform on the heath and status of the current plants.
Where we go from here
Marketers cannot avoid the sheer volume of video content emerging, whether a video company or not. Without the right tools, including CV and Active Learning, these teams are burdened with an insurmountable and draining albatross of playing “whack-a-mole” against the rise of video content with all its implications: brand safety, scalable classification, effective audience targeting…
Video is also a powerful tool outside of the marketing realm, creating new solutions for any industry where early intervention and scaling operations are critical for performance.
Our vision is to empower the industry to unlock the value within their video assets and provide relief and structured data classification, and purpose-built solutions to harness the massive amount of content created with each hour to turn it into an advantage...
We aim to provide an infrastructure for all, through which every company can build their own applications and products to start unlocking the value of their video assets.