using deep neural networks for fashion applications

50
Using Deep Neural Networks for Fashion Applications Ahmad Qamar

Upload: ahmad-qamar

Post on 03-Mar-2017

603 views

Category:

Technology


1 download

TRANSCRIPT

Page 1: Using deep neural networks for fashion applications

Using Deep Neural Networksfor Fashion Applications

Ahmad Qamar

Page 3: Using deep neural networks for fashion applications

motivation

challenges

related work

data collection

Thread Genius

applications

future work

demosthreadgenius.co

Page 4: Using deep neural networks for fashion applications

set of terms used interchangeably

● features, fingerprints, representations, latent factors, vectors

● attributes, labels, concepts

● product , SKU, in-store

● street, wild-type, UGC

threadgenius.co

Page 5: Using deep neural networks for fashion applications

threadgenius.co

motivation

Page 6: Using deep neural networks for fashion applications

threadgenius.co

2000 today→ web content increasingly image-heavy→ more participation and engagement+ 1.8B+ photos uploaded daily (2014)

- only 15% contain relevant metadata

Page 7: Using deep neural networks for fashion applications

→ millennials are increasingly brand agnostic

- logo detection fails in extracting signalthreadgenius.co

Page 8: Using deep neural networks for fashion applications

threadgenius.co

millennials prefer branded content that is ...

● social is saturated with content● attention is limited commodity

authentic and information dense

Page 9: Using deep neural networks for fashion applications

threadgenius.co

challenges

Page 10: Using deep neural networks for fashion applications

+ attribute extraction works well- limited and generic taxonomy of attributes- word attributes don’t fully capture image

threadgenius.co

visual understanding is becoming ubiquitous

Page 11: Using deep neural networks for fashion applications

- visual search suffers from poor results- lack of focus on specific domain

threadgenius.co

vs.

Page 12: Using deep neural networks for fashion applications

query out-of-the-box model fashion domain model

Page 13: Using deep neural networks for fashion applications

threadgenius.co

Page 14: Using deep neural networks for fashion applications

threadgenius.co

Page 15: Using deep neural networks for fashion applications

threadgenius.co

Page 16: Using deep neural networks for fashion applications

collaborative filtering not fit for fashion

“people who buy a also buy b”

- no information about content

- half-life of fashion products is ~1.5 months

- higher quantity (vs. movies), lower volume (vs. music)

threadgenius.co

Page 17: Using deep neural networks for fashion applications

image models require lots of data to prevent overfitting

- limited public datasets for fashion

- taxonomy requires domain consideration

- collecting training data is painstaking

- ambiguity and variability present in fashion products and photos

threadgenius.co

Page 18: Using deep neural networks for fashion applications

threadgenius.co

related work

Page 19: Using deep neural networks for fashion applications

threadgenius.co

Convnets 101: convnets learn features and the classifier simultaneously

images from Sander Dieleman

image

Page 20: Using deep neural networks for fashion applications

[Kiapour et al] Where to Buy It: Matching Street Clothing Photos in Online Shops

+ learn a NN similarity function between query and candidate to outputs match-score; works well

- querying is expensive since score must be computed for all candidates

- learn separate functions for each category (tops, footwear, …)

threadgenius.co

Page 21: Using deep neural networks for fashion applications

[Liu et al] DeepFashion: Powering Robust Clothes Recognition and Retrieval with Rich Annotations

+ provide 800K fashion images for academic community

+ perform landmark detection for localization

- questionable data quality: taxonomy contains irrelevant words (eg. brooklyn, kurt), images are mislabeled, bounding boxes too tight

threadgenius.co

images for “brooklyn”

Page 22: Using deep neural networks for fashion applications

[Bell et al] Learning visual similarity for product design with convolutional neural networks

threadgenius.co

+ learn image embeddings that place similar products close

- requires cropped images for querying

Page 23: Using deep neural networks for fashion applications

[Bell et al] Learning visual similarity for product design with convolutional neural networks

threadgenius.co

train on two tasks

● metric learning● category prediction

+

-

Page 24: Using deep neural networks for fashion applications

[Ren et al] Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

threadgenius.coobject detection model

Page 25: Using deep neural networks for fashion applications

[Ren et al] Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

threadgenius.co

train a mini-network that classifies objectness and regresses bounding boxes

Page 26: Using deep neural networks for fashion applications

threadgenius.co

data collection

Page 27: Using deep neural networks for fashion applications

fashion products exhibit different types of structure● category: top, footwear, ...● product: sweater, pumps, ...● detail: shawl collar, square toe, ...● color: yellow, anthracite, ...● pattern: paisley, colorblock, ...● aesthetic: preppy, scandalous, …

- degrees of freedom make similarity search subjective

threadgenius.co

Page 28: Using deep neural networks for fashion applications

compiled a taxonomy of ~1000 fashion attributes

threadgenius.co

Page 29: Using deep neural networks for fashion applications

for each attribute, images are sourced from● retail websites● social media networks (Pinterest, The Hunt)

● and fashion resale networks (Poshmark, Vinted)

threadgenius.co

Long sleeve silk chiffon shirt-style dress featuring graphic pattern in navy, burgundy, and green. Vented crewneck collar. Gathering at front and back yokes. Detachable self-tie fastening at waist. Two-button barrel cuffs. Detachable viscose chiffon slip lining in black. Tonal stitching. Body: 100% silk. Lining: 100% viscose. Imported.

Page 30: Using deep neural networks for fashion applications

+ image-attribute pairs help model learn fashion feature detectors- image-attribute pairs are not enough: attribute classifiers simply compute histograms over visual features

threadgenius.co

Page 31: Using deep neural networks for fashion applications

+ image-image pairs allow for unsupervised learning of similarity+ captures invariances

threadgenius.co

+ -

Page 32: Using deep neural networks for fashion applications

images require cleaning

threadgenius.coimages from [Kiapour et al]

Page 33: Using deep neural networks for fashion applications

3M+ images annotated

threadgenius.co

Page 34: Using deep neural networks for fashion applications

threadgenius.co

Thread Genius

Page 35: Using deep neural networks for fashion applications

images/UGC

unified space of images, products, and metadata

products

OPENING CEREMONY

$495

ZARA$125metadata

varsity jacket, color black,

standup collar threadgenius.co

Page 36: Using deep neural networks for fashion applications

threadgenius.co

object detector

RPN model trained on bounding box labels

feature extractor

alternate training on attribute classification and metric learning

indexing

Page 37: Using deep neural networks for fashion applications

Tech StackResearch pipeline

Deployment pipeline

Training data

...

Data annotation

Lasagne Annoy

Model training + experimentation

+ validation

Compute + Storage + Server

TG GPU Server

Product Inventory

Lookbooks

...

fashion blogs ...

Workflow manager

TG API

threadgenius.co

Page 38: Using deep neural networks for fashion applications

threadgenius.co

applications

Page 39: Using deep neural networks for fashion applications

retail

alternative products to: sold out inventory, pricey items

shop.threadgenius.co

Page 40: Using deep neural networks for fashion applications

visual marketing

making Instagram shoppable

+ 2-3x lift in conversion

threadgenius.co

Page 41: Using deep neural networks for fashion applications

audience generation

build custom audiences for specific products

threadgenius.co

Page 42: Using deep neural networks for fashion applications

audience generation

+ 3x increase in CTR

threadgenius.co

Page 43: Using deep neural networks for fashion applications

threadgenius.co

future work

Page 44: Using deep neural networks for fashion applications

threadgenius.co

experimental model [beta]

word2vec on collections of images and text

● words represented by embedding / lookup table● images represented by convnet

Long sleeve silk chiffon shirt-style dress featuring graphic pattern in navy, burgundy, and green. Vented crewneck collar. Gathering at front and back yokes. Detachable self-tie fastening at waist. Two-button barrel cuffs. Detachable viscose chiffon slip lining in black. Tonal stitching. Body: 100% silk. Lining: 100% viscose. Imported.

document

Page 45: Using deep neural networks for fashion applications

threadgenius.co

other

● combined model that directly extracts features of component apparel items (convnet+RNN)

● refine training of experimental image-text model

Page 46: Using deep neural networks for fashion applications

threadgenius.co

demos

Page 47: Using deep neural networks for fashion applications

threadgenius.co

experimental model: semantic arithmetic with images and text

Page 48: Using deep neural networks for fashion applications

threadgenius.co

robo-Bill Cunningham

Page 49: Using deep neural networks for fashion applications

Questions?

we’re hiring for backend and ML roles

Page 50: Using deep neural networks for fashion applications

we’re hiring for backend and ML roles

[email protected]