Data Collection, Image Annotation

Leveraging Computer Vision Technology for Intelligent Inventory Management

• • • •The Solution• • • •

As early adopters of this new computer vision technology, Aapya Solutions relies on feeding the model with high-quality data to expand capabilities and advance structural knowledge while prioritizing the protection of customer data as an emerging startup. When approaching DataForce, there was a need to build a data infrastructure by acquiring and annotating data at scale.

Starting near Aapya’s Tampa Bay-area headquarters, DataForce assisted in sourcing locally by building a custom data collection pipeline and quality assurance (QA) process, managing the data curation from a project management perspective.

To achieve this, DataForce enabled a full workflow through our proprietary platform, allowing for a seamless data flow from collection to annotation through delivery while maintaining data security. The full scope of work included:

• Collection of photos of aisle segments as well as their corresponding individual bottle barcodes;
• Provision of up to eight attributes per bottle;
• Segmentation of each bottle and barcode using polygons and bounding boxes, respectively.

First, we leveraged our community to find a team member who could easily travel to the indicated liquor store over a span of three days to take photos of over 300 products and their barcodes.

The challenge was how to reduce collection and annotation time, as all attributes needed to match exactly between the different data types. To enable this, DataForce created a workflow using a system of three coordinates (aisle segment + shelf number + position in shelf). The team member was able to snap the photos directly into our cloud through DataForce Contribute, our mobile application, then provide those coordinates and the corresponding attributes (from brand to size) for only one photo type, without having to repeat attributes across the two photo types. This reduced not only the time to complete the task in both collection and annotation, but also the cost to the customer—all while improving quality and eliminating risks of extensive manual data entry.

Once images were collected, we provided segmentations by way of polygons (for bottles) and square bounding boxes (for barcodes) for all the bottles across all data types.

To support this, our expert technology team built a script-based solution that was able to read the coordinates, transpose the corresponding attributes across all photo types and respective segmentations for annotation, and build scripts to generate a YoloV8 format—a first for DataForce.

The Benefits

The result was six aisle segment photos and 350 individual bottle photos, resulting in:

In addition, the customer asked us to segment another two wide-angle images of the same aisle segments and bottles, which resulted in an additional 468 segmentations with 1,404 attributes in total (three provided per segmentation in order to match initial scope).

Because of the need for high-quality data, DataForce performed sampling of 80% on collection and 84% on annotation, achieving a pass rate of 98% on collection and 97% in annotation (over 500 polygons/bound boxes and 5,000 attributes reviewed), resulting in a customer pass rate of 99.8%.

The project was completed in less than one month.

As the project progresses and the Vision-37 platform evolves with the real-time, high-quality insights it needs to operate with maximum efficiencies, Aapya’s focus will be to improve the guest experience by providing retailers results that will improve their top and bottom lines in tandem. DataForce is proud to serve as a partner to Aapya in delivering the curated data needed to provide much-needed insights on missing and misplaced products, easing inventory burdens for their clients. It is our goal to provide Aapya with the data necessary to continue refining the Vision-37 platform and processes to scale locally, regionally, and nationally.    
 

“The result of the DataForce and Aapya partnership puts us closer to winning a revenue stream from these early liquor store adapters by the end of 2023 and Q1 of 2024. We are super excited about the potential to scale this across other stores and meet our business goals by the end of 2024. We feel that Aapya in partnership with companies like DataForce can be the leading player in this space, as we have learned some key nuances with respect to computer vision, data collection, data prep, and cleansing along with the associated logistics.”

– Ramesh Tirumala, CEO, Aapya Solutions