Retail & ecommerce
Products and shelves, identified among thousands that look alike.
Product recognition, catalogue tagging and live verification for retail and ecommerce AI.
From the catalogue to the checkout
Every part of the workflow where we can take work off your team.
Catalogue
Store
Food and fresh
Search and content
We work with product images, shelf video and store CCTV.
What we label
Every class is written into a guide with your team before the first batch. Files come back as COCO JSON, YOLO, Pascal VOC, CSV, JSONL or your own schema.
Product and food images
Products
Bounding box and class
SKU, brand, variant, pack size
Food
Class
Item, ripeness, spoilage stage
Attributes
Tags
Colour, material, size, style
Shelf and store video
Shelf
Bounding box and OCR
Facing, gap, price tag, promotion
Shoppers
Track ID and time span
Path, dwell, pick up, put back
Text
Listings
Class and entities
Brand, attribute, policy violation
Search
Relevance grade
Exact, substitute, irrelevant
How we test the model
The tests that matter for retail & ecommerce models, and what you get back from each.
Look alike accuracy
Near identical items reviewed as their own set.
You get backConfusion between similar SKUs
Search relevance
Reviewers grade results for real queries.
You get backRelevance by query type
Shelf audit accuracy
Model shelf counts and gaps compared with reviewed shelf images.
You get backCount and gap error by aisle and category
Where retail & ecommerce teams start
Problems teams bring us, and the first piece of work we would scope.
Your catalogue has thousands of items that look almost the same.
What we doFine grained labels with a reference image for every look alike pair.AnnotateVolume jumps when you add stores, and quality cannot drop.
What we doWe scaled Client D twelve times in five weeks with quality held between 95 and 98%.Live reviewSearch returns the right category and the wrong product.
What we doRelevance judgments on real queries, reported by query type.Evaluate & alignRelated work
Twelve times the volume in five weeks, with quality holding at 95 to 98%
Human in the loop support for visual deviation detection on production lines, with a response time requirement under two minutes.
24/7 live review for production line visual AI, quality above 95% by month two
2,000 robot videos described every week by a team built in seven days
Half a million images for autonomous navigation, with every file kept on EU servers
Client and volumes [REAL DATA: Client names and volumes]
Bring us the retail & ecommerce data your model gets wrong.
Send a small batch and get it back done, free, usually within 24 to 48 hours. You judge the work before you pay for anything.