ParallelDots AI technology is built on the paradigm of learning more from less. Much of modern AI success comes from the abundance of data but ParallelDots technology shines in the field where annotated data is in less of supply.
Combining the best of object detection and optical character recognition (OCR), ShelfWatch achieves 95%+ product recognition accuracy at the SKU level. For some categories, we have achieved 99.5% accuracy level at steady state
ParallelDots Image Recognition solutions are powered by the same deep learning algorithms powering self-driving cars and facial recognition on smartphones. However, when it comes to product detection on the retail shelf, there are some unique challenges that we have to overcome.
On-Device Image Recognition (ODIN) is the most cutting-edge offering from the ParallelDots stable. It allows instant reporting from shelf photos captured by the field reps by processing them on their hand-held device. ODIN is fast and works completely offline. With encouraging results in the pilot stage, ParallelDots offers on-device image recognition for specific domains - low number of SKUs, limited KPIs. Read More >
Monitoring price compliance through price detection is an important feature of our AI retail image recognition solution. Here, the ShelfWatch Computer Vision detects price displays, which are read using the Optical Character Recognition (OCR) technology. The AI then attributes each SKU to the price display, and checks for compliance. Read More >
One good quality image is all it takes to setup ShelfWatch for product recognition. Deep Learning algorithms are notorious for requiring large amounts of “labeled” data to achieve high accuracy. However, collecting such a huge amount of data is very expensive for product recognition and therefore, we have created a technique to train our algorithms using just one single SKU image ("pack shots" or "hero image"). Our setup time and costs are one of the lowest in the industry.
ShelfWatch mobile app comes pre-loaded with unique features that improves photo quality at the point of capture and reduces the number of rejected images at the backend. Such features include real-time blur, glare detection and sharp angles detection in the shelf photos. All these features work in offline mode as well.
Image Recognition Solutions have traditionally been expensive to own but ShelfWatch changes the game. Other vendors require lots of data or need high-resolution shots taken in studio environments and struggle to keep the costs down for setting up Image Recognition. A lot of images are rejected in the field due to a lack of sophisticated technology to detect poor quality images in real-time. But ShelfWatch’s one sample SKU image training requirement and unique algorithms to improve photo quality at the point of capture reduces the overall costs required to deploy ShelfWatch.
ShelfWatch provides an easy-to-use and intuitive portal to visualize all the data and gain actionable insights.
ParallelDots pursues original and novel research work in the field of Computer Vision. Our research thesis is based on building applied AI solutions that can be immediately put into production.