SLIDE 1
2
Matching fashion products with image similarity Eddie Bell - - PowerPoint PPT Presentation
Matching fashion products with image similarity Eddie Bell - - PowerPoint PPT Presentation
Matching fashion products with image similarity Eddie Bell eddie@lyst.com @ejlbell 2 we collect the world of fashion into a customisable shopping experience 3 3 A python data startup that happens to be in fashion 4 What makes us di fg
SLIDE 2
SLIDE 3
3 3
we collect the world of fashion into a customisable shopping experience
SLIDE 4
A python data startup that happens to be in fashion
4
SLIDE 5
5
What makes us difgerent?
All data is scraped from retailers 400 spiders (scrapy), 9000 designers Almost everything is automated SEO, recommendation, classification, sales This architecture comes with a few problems
SLIDE 6
6
Duplicates
SLIDE 7
7
Why do we get duplicates
There is no ISBN for fashion
inter-retailer intra-retailer intra-retailer burberry selfridges
SLIDE 8
8
How we used to fjnd duplicates
Lucene fuzzy string matching yoox.com 3000 products called ‘dress’ doesn’t really work 7000 products called ‘shirt’
SLIDE 9
9
BRISK image descriptors
How we detect duplicates now
Leutenegger, Chli and Siegwart BRISK: Binary Robust Invariant Scalable Keypoints. ICCV 2011: 2548-2555.
SLIDE 10
10
How we detect duplicates now
Brisk Octaves
SLIDE 11
11
From descriptors to image similarity
k-means n x 64 n x 1 bag of words 1 x k
SLIDE 12
12
Architecture
Started in Storm / Java very painful Ended up in Celery much nicer Matching is done in elastic search
SLIDE 13
13
Architecture
SLIDE 14
Results
14
SLIDE 15
Bonus Colour variants
15
SLIDE 16
Bonus Matching sets
16
SLIDE 17
Bonus Model detection
17
SLIDE 18
Moderation
18
SLIDE 19
19
What’s next?
Reverse image search Similar textual features Dual image / text vector embeddings
SLIDE 20