This week in AI and Machine Learning: An IDC Market Note showcases Plainsight, AI supports pest detection, and more.
The ecosystem of computer vision providers and solutions is constantly evolving. It’s evolving so, quickly, in fact, that standardized techniques and approaches are only just beginning to come into view. Organizations often set unrealistic expectations and lack the resources necessary to make computer vision implementation a reality.
In their recent market note, IDC explores the computer vision landscape, discusses some of those obstacles keeping many enterprises from achieving their goals, and describes how Plainsight’s customer-focused approach has made for consistent success.
Artificial Intelligence News
AI’s First “Cryptid” Haunts Dreams and Starts Discussions
AI-generated art is having a big year, with solutions like DALL-E 2 wowing users across the globe and sparking debates about the nature of creativity and death of artistry. As Halloween approaches, digital art and AI models have reentered the discourse in appropriately spooky fashion. Twitter user @supercomposite, a musician and AI professional from Sweden, went viral after sharing a thread describing her encounters with a frightening AI-generated woman (click at your own risk!). By experimenting with an image-generation platform and negative prompts, @supercomposite first generated images of a woman with slightly surreal, unsettling features. Additional prompts inspired more and more disturbing imagery.
In addition to nightmares, Loab has inspired both skepticism and controversy. Some have pointed out the data bias evident in associating asymmetrical features and rosacea-like redness with terror. Others wonder if they’re just taking compelling bait. SuperComposite spoke to CNET to address allegations that she was merely spinning a creepy digital yarn and leaving out key details. “It’s a creepypasta,” she clarifies, “since I embellished the creepiness but the process and the phenomena are totally accurately described in my thread.”
While SuperComposite hasn’t revealed all of the prompts she used or the image-generating solution responsible for bringing Loab to life, they assure CNET readers it’s possible to encounter this “AI cryptid” at home with similar negative image prompts. Other Twitter users have taken this as an invitation to share pictures of their own. However Loab originated, she’s now all over the internet. Pictures of Loab will help to train image generation models, ensuring she’ll haunt us forever.
Creating 3D Models From Single-View Images
The human eye is capable of helping the brain imagine 3D recreations of 2D images. For familiar objects and new ones alike, a single image is typically enough. Computer vision models, however, typically require multi-view data to achieve the same level of understanding. LOLNeRF, which Google researchers presented at a recent conference, combines two existing techniques to create accurate 3D models from collections of single-view images.
Among the key resources used in LOLNeRF’s development was MediaPipe Face Mesh, which estimates facial landmarks in three dimensions based on 2D images. With just five key points, Google’s researchers were able to create a model capable of accurately estimating the facial structure of dogs, cats, and people. Read more about how the team accounted for challenges related to image clarity and brought their model to fruition.
Microscopic Pest Detection and Classification with Computer Vision
Infestations pose a major risk to the global food supply chain, necessitating quick and accurate identification to plan an effective response. Unfortunately, the quick gestation cycles and miniscule size of most pests makes manual detection a challenge.
A study, published in Agriculture, details how researchers at North Carolina State and the University of Puerto Rico leveraged the Multi-Camera Array Microscope and an automated pipeline for processing visual data to build a machine learning approach capable of categorizing caterpillar eggs with greater than 99% accuracy. Tobacco budworm and bollworm eggs look nearly identical to the naked eye. Distinguishing the difference typically requires manual reviews by highly-trained professionals with advanced technology. By first building a dataset of more than 5500 images, the researchers were able to develop a model capable of quickly and accurately labeling eggs. Farmers can now snap a photo of a leaf and instantly identify the species to intervene before they hatch.
Learn more about how Plainsight supports agribusinesses in managing pests, promoting the health and well being of livestock, and more.
Join our Community
See you next week! Until then, keep the conversation going on Plainsight’s AI Slack Channel.
About the Author & Plainsight
Bennett Glace is a B2B technology content writer and cinephile from Philadelphia. He helps Plainsight in its mission to make vision AI accessible to entire enterprise teams.
Plainsight’s vision AI platform streamlines and optimizes the full computer vision lifecycle. From project strategy, through model deployment, and ongoing monitoring, Plainsight helps customers successfully create and operationalize vision AI applications to solve highly diverse business challenges.