leveraging social media with computer vision

45
Leveraging Social Media with Computer Vision TJ Torres Data Scientist, Stitch Fix Big Data Applications in Fashion MeetUp 10/2016 Informing Recommendations in Fashion and Retail

Upload: tj-torres

Post on 21-Apr-2017

590 views

Category:

Data & Analytics


0 download

TRANSCRIPT

Page 1: Leveraging Social Media with Computer Vision

Leveraging Social Media with Computer Vision

TJ Torres Data Scientist, Stitch Fix

Big Data Applications in Fashion MeetUp 10/2016

Informing Recommendations in Fashion and Retail

Page 2: Leveraging Social Media with Computer Vision

Styling Algorithms Research

Page 3: Leveraging Social Media with Computer Vision

Styling Algorithms Research

Page 4: Leveraging Social Media with Computer Vision

Styling Algorithms Research

Page 5: Leveraging Social Media with Computer Vision

Data LabsStyling Algorithms Research

Page 6: Leveraging Social Media with Computer Vision

MOTIVATION

Inventory Scaling:

Why Recommendations?

Infeasible from an efficiency perspective to look through all inventory as it scales.

Page 7: Leveraging Social Media with Computer Vision

MOTIVATION

Inventory Scaling:

Human Ability:

Why Recommendations?

Infeasible from an efficiency perspective to look through all inventory as it scales.

Stylists can’t keep all products in their memories while trying to locate the best items for each client.

Page 8: Leveraging Social Media with Computer Vision

MOTIVATION

Inventory Scaling:

Human Ability:

Why Recommendations?

Infeasible from an efficiency perspective to look through all inventory as it scales.

Stylists can’t keep all products in their memories while trying to locate the best items for each client.

Business Success:

Aid stylists in making the best decisions to better please our clients.

Page 9: Leveraging Social Media with Computer Vision

MOTIVATIONOur goal at Stitch Fix

Total Inventory

Recommendation Algo

Stylists

Filtered Items

1 2 3 4 5

Final Items Sent

Page 10: Leveraging Social Media with Computer Vision

COMPUTER VISION

Page 11: Leveraging Social Media with Computer Vision

COMPUTER VISION

New Clients

New Clothing

Cold Start Problem

No or sparse purchasing information, so how can we supplement this?

Page 12: Leveraging Social Media with Computer Vision

COMPUTER VISION

New Clients

New Clothing

Cold Start Problem

No or sparse purchasing information, so how can we supplement this?

Perception

Fashion can be difficult to describe via text/categorization.

Many times it’s easier to show what you like.

Page 13: Leveraging Social Media with Computer Vision

TURN TO IMAGES

• Style/fashion is primarily visual.

• We wish to use images for modeling purposes.

• Heuristics for how we process image data

unknown or quite complex.

• We don’t want to have to develop image

features.

• Turn to deep learning to learn the feature

extraction.

Page 14: Leveraging Social Media with Computer Vision

OUTLINE

1. Brief Introduction to NNs

2. Deep Learning for Fashion Imagery

3. Recommendations and Social Media

4. Results

5. Conclusions

Page 15: Leveraging Social Media with Computer Vision

NEURAL NETWORKS

http://www.wired.com/2013/02/three-awesome-tools-scientists-may-use-to-map-your-brain-in-the-future/

Page 16: Leveraging Social Media with Computer Vision

http://googleresearch.blogspot.com/2015/06/inceptionism-going-deeper-into-neural.html

Page 17: Leveraging Social Media with Computer Vision

WhoaDude!

http://googleresearch.blogspot.com/2015/06/inceptionism-going-deeper-into-neural.html

Page 18: Leveraging Social Media with Computer Vision

Gatys, et. al. : https://arxiv.org/abs/1508.06576

Page 19: Leveraging Social Media with Computer Vision

Begin with input:

INTRO TO NEURAL NETS1 2 3 4 5 6

Page 20: Leveraging Social Media with Computer Vision

Begin with input: 1 2 3 4 layer 1 (Input)

5 6

layer 2

f

(l)i (x) = tanh

0

@X

j

W

(l)ij x

(l�1)j + b

(l)

1

A

INTRO TO NEURAL NETS

Page 21: Leveraging Social Media with Computer Vision

Begin with input: 1 2 3 4 layer 1 (Input)

5 6

layer 2

f

(l)i (x) = tanh

0

@X

j

W

(l)ij x

(l�1)j + b

(l)

1

A

layer 3 (output)

Transform data repeatedly with non-linear function.

f

(1) � · · · � f (n)(x)

INTRO TO NEURAL NETS

Page 22: Leveraging Social Media with Computer Vision

1 2 3 4 layer 1(Input)

5 6

layer 2

layer 3(output)

Calculate loss function and update weights

f

(1) � · · · � f (n)(x)

L(xout

, y) =

MSEz }| {1

m

mX

k=1

(xk � yk)2

Begin with input:

f

(l)i (x) = tanh

0

@X

j

W

(l)ij x

(l�1)j + b

(l)

1

A

Transform data repeatedly with non-linear function.

INTRO TO NEURAL NETS

Page 23: Leveraging Social Media with Computer Vision

1 2 3 4 layer 1(Input)

5 6

layer 2

layer 3(output)

L(xout

, y) =

MSEz }| {1

m

mX

k=1

(xk � yk)2

W (l)⇤ij = W (l)

ij

✓1� ↵

@L@Wij

Calculate loss function and update weights

f

(1) � · · · � f (n)(x)

Begin with input:

f

(l)i (x) = tanh

0

@X

j

W

(l)ij x

(l�1)j + b

(l)

1

A

Transform data repeatedly with non-linear function.

INTRO TO NEURAL NETS

Page 24: Leveraging Social Media with Computer Vision

1 2 3 4 layer 1(Input)

5 6

layer 2

layer 3(output)

L(xout

, y) =

MSEz }| {1

m

mX

k=1

(xk � yk)2

W (l)⇤ij = W (l)

ij

✓1� ↵

@L@Wij

◆@L

@W

(l)ij

=

✓@L

@x

out

◆✓@x

out

@f

(n�1)

◆· · ·

@f

(l)

@W

(l)ij

!

Calculate loss function and update weights

f

(1) � · · · � f (n)(x)

Begin with input:

f

(l)i (x) = tanh

0

@X

j

W

(l)ij x

(l�1)j + b

(l)

1

A

Transform data repeatedly with non-linear function.

INTRO TO NEURAL NETS

Page 25: Leveraging Social Media with Computer Vision

RECS AND SOCIAL MEDIA

Clients give Pinterest board to visually indicate fashion tastes.

Match pinned images to our own styles.

Page 26: Leveraging Social Media with Computer Vision

RECS AND SOCIAL MEDIA

Clients give Pinterest board to visually indicate fashion tastes.

Match pinned images to our own styles.

Strategies

Page 27: Leveraging Social Media with Computer Vision

RECS AND SOCIAL MEDIA

Clients give Pinterest board to visually indicate fashion tastes.

Match pinned images to our own styles.

Strategies

Attribute extraction and matching.

Page 28: Leveraging Social Media with Computer Vision

RECS AND SOCIAL MEDIA

Clients give Pinterest board to visually indicate fashion tastes.

Match pinned images to our own styles.

Strategies

Attribute extraction and matching. Visual feature similarity.

Page 29: Leveraging Social Media with Computer Vision

RECS AND SOCIAL MEDIA

Clients give Pinterest board to visually indicate fashion tastes.

Match pinned images to our own styles.

Strategies

Attribute extraction and matching. Visual feature similarity.

Metric learning.

Page 30: Leveraging Social Media with Computer Vision

RECS AND SOCIAL MEDIA

Clients give Pinterest board to visually indicate fashion tastes.

Match pinned images to our own styles.

Strategies

Attribute extraction and matching. Visual feature similarity.

Metric learning. …or some combination.

Page 31: Leveraging Social Media with Computer Vision

VISUAL FEATURES

Page 32: Leveraging Social Media with Computer Vision

VISUAL FEATURES

Page 33: Leveraging Social Media with Computer Vision

VISUAL FEATURES

Use pre-trained extracted features.

Compare image features with metric of your choice

Cosine Euclidean etc,

Page 34: Leveraging Social Media with Computer Vision

EXAMPLESQuery Image

Top 5 Results

Page 35: Leveraging Social Media with Computer Vision

EXAMPLESQuery Image

Top 5 Results

Page 36: Leveraging Social Media with Computer Vision

CHALLENGESQuery Image

Top 5 Results

Sometimes things don’t work out so well…

Need system to compare images across separate domains

Page 37: Leveraging Social Media with Computer Vision

METRIC LEARNING

New Metricas Objective

Anch

orPo

sitive

Neg

ative

Triplet or Contrastive Loss

https://arxiv.org/abs/1404.4661

Ltriplet(a, p, n) =1

N

NX

i=1

max {d(f(ai), f(pi))� d(f(ai), f(ni)) +m, 0}!

Page 38: Leveraging Social Media with Computer Vision

METRIC LEARNING

https://arxiv.org/abs/1511.05939

m m

Positive

Negative

Before Training After Training Before Training After Training

Page 39: Leveraging Social Media with Computer Vision

METRIC LEARNING

https://arxiv.org/abs/1511.05939

m m

Positive

Negative

Before Training After Training Before Training After Training

Learn an embedding that obeys the similarity constraints.

similarity score = d�query, inventory

Page 40: Leveraging Social Media with Computer Vision

EXAMPLES

Page 41: Leveraging Social Media with Computer Vision

EXAMPLES

Page 42: Leveraging Social Media with Computer Vision

CONCLUSIONS

1. Social media images can help make better recommendations.

a) Alleviate cold start.

b) Provide new features/data for recommendations.

2. Cross-domain image matching can be difficult, but is made easier with deep learning.

3. There’s enormous potential moving forward with this type of work.

a) Attribute labeling and trend tracking.

b) Predictive models for purchasing probability.

Page 44: Leveraging Social Media with Computer Vision

ATTRIBUTE LABELING

Page 45: Leveraging Social Media with Computer Vision

GENERATIVE FASHION