STMicro Updates Its TinyML STM32 Model Zoo, Now Offers Over 140 Ready-to-Use ML, AI Models
The new 4.0.0 release is, the company boasts, the industry's biggest collection of AI and ML models for microcontrollers.
STMicroelectronics has announced a major expansion to its tinyML "model zoo," a collection of machine learning and artificial intelligence (ML and AI) models designed to run on its STM32 family of microcontrollers — claiming that it's now the industry's largest, at over 140 total.
"Turning data science into a working application tuned for an embedded platform is a complex engineering challenge, and developers need support throughout the journey," says STMicro's Stephane Henry, vice president of the company's Edge AI Solution Group. "While expanding the selection of models available, to help the STM32 developer community jump-start their projects, we are also strengthening the infrastructure all the way to deployment with STM32 AI Model Zoo 4.0. This is part of our commitment to make Physical AI a reality."
The upgraded model zoo, provided as part of the company's ST Edge AI Suite software offering, now boasts a total of 60 model families, up from 30, which breaks down to a total of more than 140 ready-made models for a variety of real-world tasks. The update also adds native support for PyTorch models, on top of STMicro's existing support of TensorFlow Lite (TFlite) Keras, LiteRT, and ONNX formats, and offers new compressed and sub-byte four-bit quantized models for resource-constrained projects.
Among the new use-cases highlighted by STMicro in its latest release is an example of depth estimation, which attempts to estimate depth from a two-dimensional single-camera input — something the company says can be used for obstacle avoidance in low-cost robotics, as well as for mapping an environment.
The new model zoo is available in the ST Edge AI Suite now and will be released as a standalone offering on GitHub "soon," STMicro promises.