Problem: “An image is worth a 1000 Classifications”
The story of a Retail Company implementing Frictionless Checkout.
Our customer has a global presence with a catalog of thousands of products all in different languages. They couldn’t find a Data Labeling provider that could give them the global cultural context to not only identify thousands of brands and products but also have the technology to be able to mark each image in shopping carts with 99% accuracy.
Not only was this slowing down their process to get their training data labeled, but they tried to handle it in-house which resulted in an arduous, expensive effort that slowed down their roadmap substantially. They knew they had to make a change.
Solution: “A Friction-less Data Labeling Process”
By using Tasq.ai, our customer was able to receive the global labeled data they once thought impossible, labeled within hours. Turn-around time for datasets was completed in hours and days not weeks and months like before. Due to Tasq.ai’s patent-pending workflow technology, our platform deconstructed large datasets into millions of micro-tasks to be labeled by Tasqers all around the world, resulting in 99% accuracy. This revolutionized our client’s data labeling process and they are now one of the industry leaders of global frictionless checkout.
The Workflow built by our Customer Success team
- Build an intuitive multi-layered image based catalogue to ensure representation of all products with multi angle images.
- Build a task for spotting, identifying and marking every object with multiple dynamic judgements for ensuring confidence scores.
- Automatic labeling of bounding boxes around every marked object using ML for high confidence.
- Routing tasks by languages to different demographic located Tasqers.
- Human classification of every object per the catalogue, with multiple judgements to ensure high quality and confidence results.
- Segmentation by polygons of every product at pixel perfect level.
- Define rates of occlusion and orientation of angle (front, back, sideways).
- Aggregation of all results and confidence scores per annotation to one single structured image.
99% accuracy datasets for training and improving the models to support frictionless checkout retail shops with global presence. Immediate ROI at checkouts.