Monday, April 24, 2023

Sony AITRIOS wins award at tinyML 2023


At the tinyML Summit 2023, held from March 27 to 29, 2023, Sony Semiconductor Solutions' edge AI sensing platform service, AITRIOS™, won the tinyML Awards 2023 "Best Innovative Software Enablement and Tools".

The tinyML Summit is a global conference on tiny machine learning (TinyML), held since 2019, where business leaders, engineers, and researchers gather to share information on the latest TinyML technologies and applications. This year the conference was held in San Francisco, United States. This award is presented to an individual, team, or organization that has created innovative software tools or development support tools related to TinyML and has contributed to the evolution of this technology.

 Deploying Visual AI Solutions in the Retail Industry
Mark HANSON , VP of Technology and Business Innovation, Sony Semiconductor Solutions of America
An image sensor with AI-processing capability is a novel architecture that is pushing vision AI closer to the edge to enable applications at scale. Today many AI applications stall in the PoC stage and never reach commercial deployment to solve real-world problems because existing systems lack simplicity, flexibility, affordability, and commercial-grade reliability. We’ll investigate why the retail industry struggles to keep track of stock on its retail shelves while relying on retail employees to manually monitor stock and how our (AITRIOS) vision AI application for on-shelf-availability can eliminate complexity and inefficiency at scale.


The name “AITRIOS” consists of the platform keyword “AI” and “Trio S,” meaning, “three S’s.” Through AITRIOS, SSS aims to deliver the three S’s of “Solution,” “Social Value,” and “Sustainability” to the world.

Through this platform, SSS seeks to facilitate development of optimal systems, in which the edge and the cloud function in synergy, to support its partners in popularizing and expanding environmentally conscious sensing solutions using edge AI, and to deliver new value and help solve challenges faced by various industries.

AITRIOS integrates an AI model and application development environment, a marketplace, cloud-based services , and other items required for solution development into a powerful and flexible platform.

SSS, a leading company in image sensors, offers sensor configurations optimized for edge AI, enabling partners to build high-performance and reliable solutions.

AITRIOS is a one-stop B2B* (business to business) platform providing tools and environments that facilitate software and application development and system implementation.

*This service is not currently available to individual customers.


  1. This AI camera will be dead-on-arrival for several reasons:
    a) a person enters a store and tries to rob the cashier. The AI camera will produce a bounding box instead of actually describing the suspect. You need pixels from the camera to report the incident.
    b) the AI camera gives speeding tickets based on speed detection. But no one believes in the AI camera since it has no data to prove why the authorities issued the ticket in the first place.
    c) the AI camera will never learn since it has already learnt everything at the time of production and the AI camera has decided not to save the data due to "privacy concerns".

  2. I have watched Sony's presentation a few times already, and have a few questions to their VP, Mark Hanson.

    1) It is claimed that an "AI camera" is better than a streaming camera due to privacy concerns. Why so?
    If privacy is the real concern, are we not better off by banning the cameras altogether at public places as done by some cities in the US?
    Privacy and security are two sides of the same coin. We need to be safe at public spaces; safe from bad actors. We cannot expect privacy in public spaces, but at the same time we don't want unauthorized access to our private spaces.
    Law enforcement officers need to protect themselves first before protecting the people they serve from the bad actors. How can they do this if they cannot record the event?
    The ability to store the camera data and play it back helped us catch bad actors such as Boston Marathon bomber, and the police involved in George Floyd's death. How do you think "AI cameras" will help in this situation?

    2) If you are a Walmart-sized company, how can you protect your stores from theft if you don't have access to camera streams, and archived data?
    You cannot flag a customer as a potential thief if you don't have 100% evidence. Otherwise, expect lawsuits which will lead to brand destruction, revenue declines, loss of profit, and eventually business failure.

    3) The cost equation in the presentation I believe pulled out of thin air. Here is why:
    2000 stores and 2000 cameras per store = 4M cameras
    Each camera at $250 and $50 installation fee = $1.2B cost of cameras. I believe $300 per camera at this volume is going to scare the customer away.

    $20K per store for networking = $40M across the stores.
    So, networking cost = 40M/1.2B = 3.4 percent of the sensor cost. But you told this story through the lens of "streaming is bad since it requires bundles of cables".

    With a volume size of 4M, we can get an HD IP camera for $50 with installation (or lower, if we don't use Sony imaging chips!), instead of at $300.
    That will change the cost of sensors from $1.2B to $200M which is 1/6 of Sony's estimates.

    4) Having all the sensor data that is owned by the customer using our inexpensive IP cameras, we can generate additional revenues for the customers by predicting the traffic at various shelves and restocking the items accordingly. This will greatly offset the capex and opex costs.
    On the other hand, the "AI camera" has no hope of generating additional revenues since it discarded the data as it is scared of privacy, streaming, and networking.

  3. What is the point this sensor with AI? I would need anyway a MCU to control the power/connectivity. What is the advantage over any image sensor (maybe cheaper) and a MCU that can do AI like STM32N6?
    Also what is the point of this AItrios? There are already multiple tools to manage AI, fleet, etc.


All comments are moderated to avoid spam and personal attacks.