The neocognitron is a hierarchical artificial neural network proposed by Kunihiko Fukushima in 1979 for pattern recognition tasks like Japanese handwritten character recognition. It learns convolutional kernels through unsupervised learning and inspired the development of convolutional neural networks, building on earlier models by Hubel and Wiesel.
A computer vision tool that analyzes deadlift form from side-on video clips using multiple AI models. It counts reps, measures bar height, tracks hip hinge angle, and evaluates back position (straight vs. rounded) with live probability scores overlaid on the video.
JBR-001 is an open-source 3D printable desktop robot powered by Arduino UNO that enables users to explore robotics, computer vision, edge AI, and physical AI applications.
The Lightbulb Computer is a speculative device combining a projector and computer vision in a lightbulb form factor that mounts in standard sockets or portable lamps. It projects information and augmented reality overlays onto surfaces in everyday spaces, responding to voice commands and gestures to enable hands-free interaction for cooking, studying, trip planning, smart home control, and ambient displays. The design offers an alternative to smart glasses by keeping computation ambient and spatial rather than confined to phones or facial devices.
World Labs, an AI startup founded in 2024 by Fei-Fei Li, Ben Mildenhall, and Justin Johnson, is joining AMD. The company focuses on spatial intelligence and physical world reasoning, recently releasing Atlas, a model for camera view prediction that advances computer vision applications across robotics, design, and entertainment.
BlueVetaUpright is a specialized API for detecting and correcting the semantic orientation of products in photographs, achieving 99.02% accuracy on cardinal rotations and outperforming GPT-4V, Gemini, and Claude by significant margins (e.g., +15.2 percentage points against ChatGPT). The system combines multiple purpose-built components to identify correct rotation angles and estimate confidence, addressing a key gap in product-image automation pipelines where metadata-based rotation cannot detect sideways photographed items.