Sensoro - a worldwide leader in IoT and smart sensor technology - has chosen Graphcore systems to deliver the AI compute behind its latest environmental and safety monitoring systems.
Graphcore IPUs will power a range of new Sensoro solutions, designed to help towns and cities become safer, greener places to live.
AI solutions trained on data from smart sensors are emerging as an essential tool in the management of modern, urban environments as they transition to more sustainable forms of energy and transportation, while also dealing with the effects of climate change.
The Sensoro ESG (Environmental, Social and Governance) solution will use Graphcore IPU compute for a range of applications for ecological ecosystem protection, citizen safety and environmental health, animal welfare and climate emergency response, including:
- Smart fire protection: The rapid growth in the use of electric bicycles has also seen an increase in e-bike charging fires. Sensoro’s monitoring system can alert building managers to the arrival of an e-bike at their premises, allowing them to assist with safe charging procedure and facilities.
- Emergency flood prevention: Sensoro is using Graphcore IPU systems to develop early warning systems based on monitoring of urban flood risk areas, enabling better disaster preparedness and emergency response.
- Smart ecological governance: Illegal river fishing breaks the ecological food chain and destroys the integrity of river ecosystems. Using IPU systems, Sensoro is able to recognise different fishing scenarios, such as electric fishing and net fishing, providing accurate information on illegal fishing activity.
Sensoro founder and CEO Tony Zhao said: "Graphcore's IPU systems provide an efficient and easy-to-use computing platform for our urban ESG solution, and solve the computing power bottleneck that we have faced for a long time.
“The efficiency of Graphcore IPUs and Poplar’s ease of use accelerates the development, optimization and deployment of multiple AI models, and gives us a computing foundation from which we can explore further innovative AI solutions.
“We will be continuing to work with Graphcore, and using the IPU to deliver positive change in more aspects of people’s lives.”
Object Detection with YOLO
One of the main AI models used by Sensoro, running on Graphcore IPUs, is YOLO (You Only Look Once), a highly effective convolutional neural network for real-time object detection. Since the release of the first version of YOLO in 2015, it has undergone a number of refinements to improve speed and accuracy.
Sensoro found that the IPU’s fine-grained and highly parallel compute capabilities lent themselves perfectly to the parallelisation required to get the most out of YOLO.
When running inference on high resolution images (1920x1080), Sensoro saw a 4x performance gain compared to the GPU-based inference solution they had been using previously.
For more detail on Sensoro’s implementation of YOLO on the IPU, we have published a technical deep-dive blog.
