Quick introduction for artificial intelligence / deep learning applications in fashion, beauty and creative industries.
Alternative download link: https://dl.dropboxusercontent.com/u/6757026/slideShare/creativeAI.pdf
OpenShift Commons Paris - Choose Your Own Observability Adventure
Artificial Intelligence in Fashion, Beauty and related Creative industries
3. Artificial Intelligence in Creative Industries
Artist and researcher Terence Broad is working on his master's at
Goldsmith's computing department; his dissertation involved
training neural networks to "autoencode" movies they've been fed.
boingboing.net/2016/06/02
Jürgen Schmidhuber, Point Omega,
https://youtu.be/KQ35zNlyG-o
http://dx.doi.org/10.1016/S0004-3702(98)00055-1
For an excellent review of the current state, see the
post by Kyle McDonald who is an artist working with
code: medium.com/@kcimc/a-return-to-machine-learning
Unsupervised Representation Lea
rning with Deep Convolutional G
enerative Adversarial Networks
Semantic Style Transfer and T
urning Two-Bit Doodles into F
ine Artworks
Deep Dream FBO Glitch By
KyleMcDonald , also
posted to Twitter.
London, United Kingdom, Sept 21 2016
Over the past few months, there has been increasing interest in applying the latest
developments in artificial intelligence to creative projects in art, music, film,
theatre and beyond.
techinsider.io, Magenta group
introduced at Moogfest
https://vimeo.com/169779284
4. Fashion & Artificial Intelligence
If artificial intelligence has its way, discounting could disappear, thanks to software that tells
retailers exactly what and how many products to buy, and when to put them on sale to sell them at full
price. Online shopping could become a conversation, where the shopper describes the dress of their
dreams, and, in seconds, an AI-powered search engine tracks down the closest match. Designers,
merchandisers and buyers could all work alongside AI, to predict what customers want to wear, before
they even know themselves.
For fashion, some of the biggest opportunities are in aligning supply and demand, scaling personal
customer service, and assisting designers.
By analysing large amounts of data — say, the browsing and shopping history of every single one of a
fashion brand’s online customers, as well as those of its competitors — AI can tell a retailer how to align
product drops to match demand, and even how to display products in a store to sell as many as
possible.
Machine learning can also enable brands to finely personalise their offerings to each market, or even,
each individual customer. IBM's Watson — which is working with over 500 partners in industries
including retail — has partnered with The North Face to offer “guided shopping” online. The AI asks
shoppers questions on factors such as gender, time of year and technical product details, to deliver
tailored recommendations.
"There are AI systems today that compose music, write stories, and create artwork that no one can tell is
machine-generated. So fashion design is surely not beyond AI's capabilities,” In the same way that the
work of architects like Frank Gehry and Zaha Hadid relies on computer modelling, “Fashion designers
armed with AIs will be similarly able to come up with radical new ideas: AI will amplify their
creativity rather than replace it," reasons Domingos.
“AI will absolutely challenge and replace designers,” counters Kenneth Cukier. “Let's get
real — lots of design is trial and error or boring, repetitive work. AI can help with both by making more
accurate predictions of what designs will work and taking over some of the repetitive tasks.”
Others agree that, for the moment, partnering with third party AI specialists is the way forward. “The
smartest thing a business can do, is partner with a fashion-focused tech company with AI at its core,”
says Geoff Watts of Edited. “Building AI teams from scratch, or acquiring AI start-ups and retrofitting
them to have a retail focus, requires a substantial investment of time and money.”
businessoffashion.com/articles/fashion-tech
www.technologyreview.com
Jonathan Zornow, the sole employee of a new startup called Sewbo, thinks the U.S. could
bring garment manufacturing a little closer to home by automating the feeding of fabric
into sewing machines—a step that to this day is done by hand. Zornow has created a
process by which a robotic arm guides chemically stiffened pieces of fabric through a
commercial sewing machine.
Apparel companies often move their manufacturing to countries where wages are
lower in a perpetual quest to cut costs. The Center for American Progress
found that in 2011, 15 of the top apparel exporters to the U.S. paid their Chinese
garment workers an average monthly real wage of $324.90. Bangladeshi workers
earned just $91.45. Meanwhile, U.S. sewing machine operators earn an average monthly
wage of $1,922, according to the Bureau of Labor Statistics.
Part of its selling point is that Sewbo users can design and begin mass producing a new
garment style in a day, as opposed to the months it typically takes to manufacture and
ship a new garment design. Such a feat would certainly bring new meaning to the term
“fast fashion.”
5. Fashion design Deep learning for design
Created in partnership with U.K.-based digital design studio Stinkdigital, Project Muze constitutes a
specially built engine that uses a neural network trained with the design preferences (color, texture, and
style) of more than 600 fashion “trendsetters” and features data from the Google Fashion Trend Report,
in addition to styles that have trended on Zalando itself.
http://www.stinkdigital.com/work/zalando-project-muze
venturebeat.com/2016/09/02
We started Project Creaite last winter to see the capabilities of deep learning and computer
vision algorithms in creative work such as fashion design and product design. The first version was
focused on creating product pictures for items from fashion vertical such as apparel and accessories.
We finally got some time to write about it and share what we did through a blog post here.
As we all know, Artificial Intelligence is finally here in its narrow form and ready to be helpful to
solve specific problems for which we have readily available data. At Artifacia Research we have
spent some time studying in detail about generative models. In phase one of Project Creaite, we used
an encoder-decoder scheme to get promising results and built a prototype a few months ago that can
come up with new designs for fashion products after being fed with enough number of examples.
We at Artifacia Research believe that the further development of this technology can help bring a lot
of efficiency in fashion design in particular and product design in general. We are really excited
by the possibilities of Deep Learning and AI and so we would be investing our time to take this project
to the next level very soon.
research.artifacia.com
Example results from our generative network“Using data from so many different sources in such a complex algorithm is not going to produce something exact, intricate, or least of all,
conforming to design norms. As you can see in the pictures attached below, the machine tends to spit out some seriously strange
designs, but with an element of cool and human to them because of their inspirations and origins. An option is presented to answer
more questions and further customize a given design, as well; this option matches the user up with a whole new pattern, style, or
shape, and shows how that trait would apply to the outfit on show. While some of these are downright impractical and most of them
would get you more than a few strange looks in public, the fact that a machine is capable of this sort of thing at all speaks volumes
about the hard work that many talented people have put into the field of machine learning down the years.” - androidheadlines.com
6. Fashion manufacturing Local customized high fashion?
ROBOTIC MANUFACTURING
Locally near the customer (let it be Tokyo, London or New York)
Make It luxurious and unique
No need for child labor in Asia, and only raw textiles need to be shipped
REFERENCES
Scanning instead of trying-on: Custom-made clothes with 3D Laser Scanning (LMS400 from SICK)
Explore Cornell - The 3D Body Scanner - Made-to-Measure
Movie: 3D scanning used to create unique fitted clothes - Dezeen
The DittoForm – A 3D Body Scan Dressform « bits of thread sewing studio
Volumental and Their 3D Scanning Technology is Bringing Custom Fit Shoes to the US
9. Fashion shopping Shopping Assistants #3
http://www.racked.com/2014/3/19/
https://indico.io/blog/fashion-matching-tutorial/
Deep Learning Opportunities
in Fashion Deep Learning is the
buzz word in Artificial Intelligence these
days. But what is it all about?
www.picalike.com/blog/2014/02/20
http://multithreaded.stitchfix.com/blog/2015/09/17/deep-style/
https://arxiv.org/abs/1609.07859
http://dx.doi.org/10.1007/978-3-319-10590-1_31
Cited by 40
10. Fashion stylist Augmenting the human?
… me sharing not only obvious information like my size, desired price range and
“daringness” (with “daring” defined as wearing floral shirts or shorts with blazers),
but also helping her work out my actual style preferences by telling her brands I
like and flicking through endless pictures of well-dressed men to highlight the looks I
want.
This is no AI horror story, though. My stylist Sophie Bailey-Hine is very real, and her
and her colleagues at Thread, a British startup that was founded in 2012, are
currently helping 480,000 men find a new image, dress well, or simply sort out their
clothes shopping.
“All through, my goal with this business is not really to build a niche quirky retail
thing. The majority of men want to dress decently, and don’t particularly love
shopping that much. So if you can find a way to spend way less time on it, with a
much much better result, I think it’s what the majority of men would use. And so I see
this as building a new default for how the majority of men buy clothes.”
The startup won a position in the prestigious Y Combinator accelerator in San
Francisco, a sort of three-month boot-camp for startups, which led to one of its first
paying customers being Instagram CEO Kevin Systrom. But after graduating, it
moved from San Francisco back to London, and after a stint in trendy Shoreditch is
now further east in a decidedly less cool part of Whitechapel.
From day one, the stylists were working as much to teach the algorithm how to aid them
as they were to get the best clothes to customers. …Now, the algorithm works better,
letting a stylist lay out the skeleton of an outfit while filling in the specifics and targeting
who actually receives it as part of their picks.
It also allows for the stylist’s own taste to filter through. “I’m obsessed with simple
Scandinavian clothes,” Bailey-Hine tells me, “while my fellow stylist Sam is big into street
style and Freddie is our very own suiting expert – everyone has their own thing.”
In a world where fears of robots “taking jobs” is rife, Thread offers some hope. Yes,
eight stylists are doing the jobs of hundreds or thousands in a pre-AI age; but
those eight are working with AI, not for it. And besides, Sophie says, “one of the best
things I can do as a stylist is to look at photos of a guy and within a few seconds think,
‘OK, he’d look better if he was wearing a slimmer jean, he could make that navy suit look
more interesting with a knit tie, and he’d look really good in a light Private White V.C
jacket for a sharper take on his utilitarian vibe.’https://www.thread.com/
theguardian.com/technology/2016/jul/19/
11. Makeup 3D print your pigments and makeup
Picture this: you snap a photo of
a tropical purple flower and in
less than two minutes you can
print a lipstick in that exact color.
And you’ll do it at home with
Mink, your personal 3D printer.
https://makeuphacker.myshopify.com/
forbes.com
https://www.foreo.com/institute/moda/
While waiting for Mink, women may have yet
another incredibly innovate option when it
comes to makeup, and yes it also uses
3D printing technology. Stockholm, Sweden-
based skincare company, Foreo, has just
unveiled MODA, ‘the world’s first digital
makeup artist’.
Digital makeup artistry
12. Makeup Algorithmic analysis and generation
https://arxiv.org/abs/1604.07102
Our system has two functions. I: recommend the most suitable makeup for each before-
makeup face II: transfer the foundation, eye shadow and lip gloss from the reference to
the before-makeup face. The lightness of the makeup can be tuned. Pay special attention
to eye shadow, lip gloss and foundation transfer.
Publication number: WO2015127394 A1
Publication type: Application
Application number: PCT/US2015/017155
Publication date: Aug 27, 2015
Filing date: Feb 23, 2015
Priority date: Feb 23, 2014
Inventors: Yun Fu, Shuyang Wang
Applicant: Northeastern University
Export Citation: BiBTeX, EndNote, RefMan
Patent Citations (5), Non-Patent Citations (3),
Classifications (7),Legal Events (2)
External Links: Patentscope, Espacenet
https://www.google.com/patents/WO2015127394A1?cl=en
13. Physical Product Design Parametric AI augmentation
http://www.getlittlebird.com/blog/3d_printing_plus_artificial_intelligence
Part of a 30 day series on the intersection of AI, and a wide
variety of other exponential technologies as explored by cross-
over influencers and experts.3D printing has the potential to
massively democratize access to manufacturing - but what if it went
even further than puting fabrication into the hands of all people?
Imagine 3D printing in the hands of machine intelligence. Artificial
intelligence plus 3D printing could yield some really transformative
experiences. Who’s paying the most attention to that intersection?
architectmagazine.com
Daedalus Pavilion is Ai Build's latest construction, 3D printed in 3 weeks by
industrial robots, using the latest technology in artificial intelligence, deep
learning, computer vision and robotics. Ai Build teamed up with partners
NVIDIA, Arup, KUKA Robotics and Formfutura to create Daedalus Pavilion as
part of the GPU Technology Conference in Amsterdam. pinterest.com
#next_top_architects #roboticfabrication #newbaby
#fibrousthread #arduino nexttoparchitects.orgdezeen.com/2015/08/26
Designer and researcher Neri Oxman and her
Mediated Matter group atMIT Media Lab
PARAMETRIC 'PORN'
Iris Van Herpen, Neri Oxman, Ana Rajcevic Studio
14. Jewelry Design Easy target for AI optimization
Ontic Design developed 3D-printed silver and gold bracelets that consumers
can design for themselves. You can go onto the website, choose the design
and create a piece of art that's coloured onto a glass plate sliding onto the
bracelet. It's one of the world's first print-on-demand jewellery companies
and the products are beautiful.
Artificial intelligence techniques are reviewed in intentions of developing algorithmic designs, and
assisting hard computations and parameter optimizations in designing and casting. With those
computer advanced technologies, major challenges and solutions are systematically discussed and
commented to address new developments for jewelry industry.
In general, product design process begins with identifying
customer needs, concept generation, concept selection,
evaluation, and prototyping. Several AI techniques have
been applied to assist designers in almost all stages of
product design. For instances, customer needs was
identified using the concepts of Quality Function
Deployment (QFD) and fuzzy set theory and fuzzy
inference. Product form was automatically created using
EA, fuzzy logic, hybrid of fuzzy, neural network and
genetic algorithms (GA), and hybrid of neural network and
kansei expert system. Fuzzy logic technique was also
applied to evaluate and select design concepts. Affective
user satisfaction was analyzed and modeled using a
fuzzy rule-based method.
In production, several researches had been applied AI
techniques for optimizing process parameters or predicting
some consideration properties, which can be applied the
concepts to jewelry casting such as the attempt that used
GA and ANN in porosity minimization of aluminum casting,
the application of GA in shape and process parameter
optimization, the optimization of product quality of casts
using heuristic search techniques and by using GA and
knowledge base.
https://onticdesign.com/en/about
http://dx.doi.org/10.1145/2807442.2807448
http://web2.eng.nu.ac.th/nuej/file/journal/NUEJ_Vol6_1_2011_paper06.pdf
15. Web site design And beyond Adobe Creative Suite
TheGrid.io the first artificially intelligent website-
design tool
Upload photographs that capture the mood and
feeling of your brand and The Grid matches them.
The Grid tries to understand your website design
needs through algorithms and the results are
promising. Tell it you want an eCommerce store, blog
and testimonials and a unique website is put together
with fonts and colours based on uploaded
photographs and more.
How can designers adapt?
1) Join the AI party Change your skillset and learn to code. Develop your own The Grid and become a
platform enabling consumers to design things for themselves.
2) Diversify your design skills to survive. Digital design which has strong structure-based frameworks will be
the first to succumb to AI's computer might: website design, digital publishing, app design... We're even seeing AI
being applied to TV show openings. Diversify towards work that requires human empathy – the one weakness
of computers – and creative storytelling to survive. Fashion films shine a light here.
In case these changes to design sound familiar to you, you're right, they are. Before Adobe released its creative
suite in 2003 digital artists had to programme their own effects, filters and templates. Motion graphics stem
from the demoscene of the eighties where groups of hackers would crack software then put a little graphical
intro at the beginning to show it was cracked. These graphics led to groups of mostly males – FullScream
included – competing to see who could make the best graphics using the limited computing power available;
these were called Demos.
And when Adobe started bringing out its software, oh boy, did people moan. Can you imagine it, people
complaining about the advent of Illustrator, Photoshop and After Effects? People wouldn't have to learn to
code anymore – the core skill then of digital creativity – and so bland artwork that all looks the same was
feared.
But what happened instead? A huge explosion in the quality, quantity and diversity of design. Millions more
people were given the toolset to make their ideas reality. It created a new industry and tens of thousands of
new jobs. The same will happen with the AI revolution. An industry will be created around developing AI
applications and digital designers will have to re-learn how to code. The movement will also spawn pseudo-AI
digital art with the decisions made being a combination of computer and human thoughts.
For example, a computer program could analyse the millions of photos on Instagram and understand what
trends and styles of imagery are gaining traction. It could choose colours, filters and effects that have the most
interaction within target markets. A draft picture could be created and then you, humble human designer, could
use it as a starting point for your own work.
www.digitalartsonline.co.uk
https://www.youtube.com/watch?v=OXA4-5x31V0
https://www.youtube.com/watch?v=tdWs_ZJL2c8 https://www.youtube.com/watch?v=Tda7jCwvSzg
http://prisma-ai.com/
https://www.youtube.com/watch?v=9c4z6YsBGQ0
DEMOSCENE PHOTOSHOPDEMOSCENE
AI FILTER
AI TOOL
16. Videography Generative models everywhere
Magic Pony Technology, created by graduates of
Imperial College London with expertise in statistics,
computer vision, and neuroscience, trains large
neural networks to process visual information.
technologyreview.com
techcrunch.com/2016/06/20/
Everyone can shoot video but few can record or afford a legal soundtrack. Until now. With
Jukedeck’s new artificial intelligence music composition technology, creators can get a cheap, royalty free
soundtrack custom-made for their video. Jukedeck users don’t even need musical talent. They just select the
mood, style, tempo and length, and Jukedeck returns a unique song to match their short film, YouTube
series or 6-second Vine.
http://web.mit.edu/vondrick/tinyvideo/
17. Virtual and Augmented Reality Low-hanging fruit
Although it looks like a real woman, the ultra-lifelike figure in this music video
is actually a digital model created using high-resolution 3D scans. London-
based multimedia studios Marshmallow Laser Feast and Analog teamed up to
create the video, called Memex, which is currently on show as part of the
Istanbul Design Biennial 2016.
https://www.youtube.com/watch?v=ALc4xoy3-Yk
variety.com/2016
dezeen.com/2016/10/21
dailymail.co.uk/sciencetech
Image Courtesy:- http://www.vrguru.com
thebimhub.com/2015/07/28, BIM, AEC, Autodesk
augment.com/augmented-reality-architecture
Image-guided medical procedures
spl.harvard.edufortune.com/2016/04/12
19. Photography Automatic curation
Posted on February 29, 2016 by Appu Shaji
3 Comments Tagged cuDNN, Deep Learning,
Machine Learning, photography, Theano,
Torch
EyeEm is a community and marketplace for passionate photographers. More than 15 million
photographers use EyeEm to share their photos, connect with other photographers,
improve their skills through masterclasses, get recognition through our photography missions and
exhibitions, andearn money by licensing their photos. The following video shows the impact of our
deep-learning-based automatic aesthetic curation on the EyeEm search experience—read on
to learn more about how it is done.
https://vimeo.com/154364175#at=55
At EyeEm we develop technology that helps photographers tell their stories and get
discovered. We believe there are two ingredients that contribute to the success of a
photograph: 1) the story behind the photograph, and 2) the way that story is told.
Automatic image tagging technologies from companies like EyeEm,
Google Cloud Vision, Flickr, and Clarifai are quickly approaching maturity and
helping to tell the stories behind photographs by indexing or tagging them to make
them discoverable.
Visual aesthetics addresses the way each story is told; specifically, how the visual
style and composition create an emotional connection with the viewer by using
structure and cues that draw attention toward (and away) from the constituent story
elements of the photographer’s choice.
Our objective is to compare images and learn the commonalities between well-
crafted photographs, and the differences between well-crafted and mediocre
photos. A loss function that can express this is as
an extension of the hinge loss function, Researchers have explored such loss functions
in image similarity, metric learning and face identification settings with great
success in the past (Chechik2010 , Norouzi2011, Schroff2015). In a larger context,
such energy functions fall into the class of implicit regression (LeCun06), where the
loss function penalizes the constraint that input variables must satisfy.
As a photography-first company, we validate
our assumptions with our internal curators
and reviewers, who spend the major part of
their working day curating photographs. We
track the time they spend on curation tasks
where the image list is prioritized via the
aesthetic rank vs. the default sort order.
Using deep learning, we have been able to
reduce curation time by 80%.
[Chechik 2010] G. Chechik, V. Sharma, U. Shalit, S. Bengio, Large Scale Online Learnings of Image Similarity Through
Ranking, Journal of Machine Learing Research 11, 2010
[Norouzi 2011] M. Norouzi, D. Fleet, Minimal Loss Hashing for Compact Binary Codes, International Conference in
Machine Learning (ICML), 2011.
[Schroff 2015] F. Schroff, D. Kalenichenko, J. Philbin , FaceNet: A Unified Embedding for Face Recognition and
Clustering, ArXiV : 1503.03832, 2015
[LeCun 2006] Y. LeCun, S. Chopra, R. Hadsell, M. Ranazato, F. J. Huang, A Tutorial on Energy-Based Learning,
Predicting Structured Data, 2006
20. Photography/Imaging → Generative Photorealistic CGI
https://www.youtube.com/watch?v=R1-Ef54uTeU
theverge.com/2014/8/29
http://www.cgsociety.org/index.php/cgsfeatures/cgsfeaturespecial/building_3d_with_ikea
So where are the deep learning / algorithms putting CG artists out of business?
These approaches still require manual work
https://blogs.nvidia.com/blog/2015/08/11/photorealistic/
This manuscript describes recently developed
technologies for better handling of image
information: photorealistic visualization of medical
images with Cinematic Rendering, artificial agents
for in-depth image understanding, support for
minimally invasive procedures, and patient-specific
computational models with enhanced predictive
power.
“From physical model to beautiful render”
https://arxiv.org/abs/1605.02029
21. Photography Smart imaging (of course)
AI Enabled vs Phase Detect Autofocusing by Jafaar Almusaad
http://cognitionx.com/ai-better-photography/
The Frankencamera F3, an experimental computational photography
camera platform. The camera runs Linux, and its metering, focusing,
demosaicing, denoising, white balancing, and other processing is
programmable. https://graphics.stanford.edu/projects/camera-2.0/
Non-AI based insipration in 'smart cameras'
Frankencamera F2
http://prolost.com/blog/lightl16
https://light.co/
Manuscripts are solicited to address a wide
range of topics on computer vision techniques
and applications focusing on computational
photography tasks, including but not limited to
the following:
●
Advanced image processing
●
Computational cameras
●
Computational illumination
●
Computational optics
●
High-performance imaging
●
Multiple images and camera arrays
●
Sensor and illumination hardware
●
Scientific imaging and videography
●
Organizing and exploiting photo/video
collections
●
Vision for graphics
●
Graphics for vision
wikicfp.com
www.ximea.com
Smart cameras with GPU enhancement
artefactgroup.com
22. Floral Business Disruptive business models
http://time.com/4117334/valentines-day-flower-delivery-startups/
http://fortune.com/2016/02/03/bloomthat-nationwide-delivery/
For a few years, flower delivery startups have emerged to try and grab a share of the $4 billion online flower industry.
No generative approaches to design floral bouquets or something similar imho. Only image recognition mainly for AgTech purposes.
They packed up and moved to San Francisco, and were accepted into the startup bootcamp Y Combinator's 2013 class.
They named the company they created BloomThat, and set out to create bouquets for same-day delivery that are more
lovely and seasonal than even local florists could pull off.
Along with several other startups in major metros around the country, Farmgirl Flowers is set on
reinventing the roughly $10 billion business of buying flowers. “Everyone says the flower industry
is a dying industry because flower shops don’t work with the overhead, and the flower companies
don’t offer what people want,” says Stembel, the company’s sole founder. “I just saw this and
thought this is absurd. How has nobody done anything in this industry?”