A.I. in the Enterprise: Computer Vision
1 like177 views
This document discusses computer vision and its capabilities. It notes that computer vision is ready for mainstream use, evolving faster than expected, and a gateway for AI applications. Computer vision allows extracting useful information from visual data, and examples of its applications include vehicle navigation, farming, visual search, and monitoring brand exposure. It is changing everything by unlocking the visual world for computers. Computer vision also creates richer experiences by understanding facial expressions and emotions. Getting started with computer vision is easier than expected using commercial cloud APIs that require no training data.
![AND COMING SOON...CREATE
Write a text description of an
existing image.
Given a text description,
generate an image from scratch.
petapixel.com/2016/09/23/googles-image-captioning-ai
-can-describe-photos-94-accuracy youtu.be/rAbhypxs1qQ?t=5s
[ 2016 ]](https://image.slidesharecdn.com/session3computervision-180710075759/85/A-I-in-the-Enterprise-Computer-Vision-19-320.jpg)
![PLUS...
[ 2017 ]
Fotogenerator.npocloud.nl
...and A.I. finishes
it for you.
Sketch a wireframe](https://image.slidesharecdn.com/session3computervision-180710075759/85/A-I-in-the-Enterprise-Computer-Vision-20-320.jpg)
A.I. in the Enterprise: Computer Vision
- 1. Extracting, harnessing and generating visual information. COMPUTER VISION
- 2. ● It’s ready for prime time ● It’s evolving faster than you think ● It’s a gateway capability ● It’s changing everything ● It creates a richer experience ● It’s easier to use than you think COMPUTER VISION O U R JO U R N EY
- 3. IT’S READY FOR PRIME TIME
- 4. And what better way to demonstrate that it’s ready for prime time, than to show it live... PS — this is an open source application written by IBM
- 6. visual-recognition-demo.mybluemix.net Photo: youtube.com/watch?v=tE2ihx0nJMI Three human faces identified Attempt to guess ages and gender of identified faces List of objects and colors in the image and the scene context Hierarchical breakdown of the scene context
- 7. CREATE A CUSTOM MODEL visual-recognition-demo.mybluemix.net/train
- 8. Classifier Application HOW IT WORKS Face, object & scene recognition Cloud Visual API Production Images
- 9. PRICING ●FEATURE 1 - 1000 UNITS/MO 1001- 1,000,000 UNITS/MO 1,000,001 to 5,000,000 UNITS/MO 5,000,001 - 20,000,000 UNITS/MO Label Detection Free $1.50 / 1000 units $1.50 / 1000 units $1.00 / 1000 units OCR Free $1.50 / 1000 units $1.50 / 1000 units $0.60 / 1000 units Explicit Content Detection Free Free with Label Detection* Facial Detection Free $1.50 / 1000 units $1.50 / 1000 units $0.60 / 1000 units Landmark Detection Free $1.50 / 1000 units $1.50 / 1000 units $0.60 / 1000 units Logo Detection Free $1.50 / 1000 units $1.50 / 1000 units $0.60 / 1000 units Image Properties Free $1.50 / 1000 units $1.50 / 1000 units $0.60 / 1000 units
- 11. API RESPONSE { "responses": [ { "labelAnnotations": [ { "mid": "/m/01yrx", "description": "cat", "score": 0.92562944 }, { "mid": "/m/04rky", "description": "mammal", "score": 0.90815818 }, { "mid": "/m/01l7qd", "description": "whiskers", "score": 0.79939437 }, { "mid": "/m/07k6w8", "description": "small to medium sized cats", "score": 0.66373962 }, { "mid": "/m/0307l",
- 12. IMAGE SIZING VISION API FEATURE RECOMMENDED SIZE NOTES FACE_DETECTION 1600 x 1200 Distance between eyes is most important LANDMARK_DETECTION 640 x 480 LOGO_DETECTION 640 x 480 LABEL_DETECTION 640 x 480 TEXT_DETECTION 1024 x 768 OCR requires more resolution to detect characters SAFE_SEARCH_DETECTION 640 x 480 cloud.google.com/vision/docs/supported-files
- 13. WHERE IS THE A.I.? In the system’s ability to determine objects and context within each image. Compared to having people view and judge images manually. Remember: A.I. is subjective
- 14. WHAT IS COMPUTER VISION? And just to clarify… Visual recognition — exploring face detection, emotion recognition, text extraction, damage identification, context awareness, and more.
- 15. EXAMPLES IN USE TODAY Manufacturing ● Ensure products are positioned correctly on an assembly line Visual auditing ● Monitor for compliance or deterioration in fleet of trucks, planes, or windmills ● Train classifiers to understand what defects look like Insurance ● Quickly process claims by classifying images of claims Social listening ● Track buzz about your company on social media Security ● Monitor for activity, instantly classify objects as threat or not Social commerce ● Use an image of a food dish to find out which restaurant serves it and find reviews ● Use a travel photo to find vacation suggestions based on similar experiences ● Use a house image to find similar homes that are for sale Retail ● Take a photo of a favorite outfit to find stores with those clothes in stock or on sale ● Use a travel image to find retail suggestions in that area Education ● Create image-based applications to educate about taxonomies ● Use pictures to find educational material on similar subjects R ESO U R C ES
- 16. Computers have already matched — and exceeded — human capabilities when it comes to understanding visual information. Now it’s just a question of finding uses and applying this new capability in productive ways. READY FOR PRIME TIME R EC A P
- 17. IT’S EVOLVING FASTER THAN YOU THINK
- 18. ALREADY BETTER THAN HUMANS February, 2015 — “...researchers say their system achieved a 4.94% error rate... In previous experiments, humans have achieved an estimated 5.1% error rate.” microsoft.com/en-us/research/blog/microsoft-researchers-algorithm-sets-imagenet-challenge-m ilestone
- 19. AND COMING SOON...CREATE Write a text description of an existing image. Given a text description, generate an image from scratch. petapixel.com/2016/09/23/googles-image-captioning-ai -can-describe-photos-94-accuracy youtu.be/rAbhypxs1qQ?t=5s [ 2016 ]
- 20. PLUS... [ 2017 ] Fotogenerator.npocloud.nl ...and A.I. finishes it for you. Sketch a wireframe
- 21. Computer vision is already better than humans at identifying objects within images. And they just started learning how to create images from scratch...and they’re already pretty good at it. FASTER THAN YOU THINK R EC A P
- 22. IT’S A GATEWAY CAPABILITY
- 23. If you’re looking for a place to begin your A.I. journey, this is a good starting point...
- 24. THE PROBLEM We live in a visual world, yet capturing useful information from images has historically required human vision — which can be slow and costly.
- 25. THE GOAL But if we could extract that useful information through computer vision, it could provide invaluable insight for business.
- 26. THE SOLUTION An intelligent visual recognition service that automatically analyzes and identifies objects and scenes in image files (video, etc.).
- 27. A HIGH PROFILE EXAMPLE Facial recognition systems are coming on strong and being used in a wide variety of applications. youtu.be/K4u4Dpl6NKk?t=1m9s
- 28. Visual recognition is 1 of 2 critical capabilities that will allow artificial intelligence to integrate into and empower our world in ways we’ve only dreamed of. It allows computers to interact with humans on our own terms — to integrate into our daily lives. GATEWAY CAPABILITY R EC A P Speech is the other one.
- 30. We’re unlocking the entire visual world for computers. What can they do with it?
- 35. INFRINGEMENT MONITORING Keep an eye out for intellectual property infringement. Photo credit: flickr.com/photos/inthe-arena/12939586573
- 36. MONITORING BRAND EXPOSURE Track how often your advertising is shown. Photo:sportvision.com/baseball/virtual-advertisements
- 37. IT’S EVEN CHANGING ‘PEOPLE WATCHING’ medium.com/homeland-security/no-longer-just-another-face-in-the-crowd-15e1c74fe24
- 38. Giving computers access to the visual world will empower our work and lives. Amplifying human productivity in endless ways. IT’S CHANGING EVERYTHING R EC A P
- 39. IT CREATES A RICHER EXPERIENCE
- 40. “Customer experience is the new battlefield.” ~ Gartner, 2015 accenture.com/us-en/insight-artificial-intelligence-ui EXPERIENCE ABOVE ALL
- 41. VIP TREATMENT A top customer walks into your store. The system instantly recognizes them, issues a personalized greeting, and alerts an attendant. And/or… Allows you to track store visits just as we track web visits in Google Analytics (faces vs. IP addresses).
- 43. EMOTION RECOGNITION popularmechanics.com/technology/a18636/facewatch-facial-recognition-identify-criminals Analyze how people react to your product. Or dynamically adjust a system based on people’s emotion.
- 44. A RICHER EXPERIENCE Computers can now see, read — and act upon — the full spectrum of our communication. Text Only Text + Speech Text + Speech + Vision Adding... Voice tone Voice inflection Adding... Facial expressions Body language Language sentiment analysis en.wikipedia.org/wiki/Albert_Mehrabian 7% 38% 55% R EC A P
- 45. It’s often said that the verbal and audible elements of communication only make up 45% of what is being said. A RICHER EXPERIENCE
- 46. IT’S EASIER THAN YOU THINK
- 47. PICK YOUR STARTING POINT A.I. M aturity No training data required! Purpose-Built Platform, Their Training Data Commercial Platform, Their Training Data Commercial Platform, Your Training Data In-House Platform, Your Training Data
- 48. NOTHING NEW HERE Most of the major cloud providers have purpose-built A.I.-powered APIs. Many have a free tier. And they work exactly like every other API you’re already using.
- 49. R ESO U R C ESLOTS OF COMMERCIAL OPTIONS Jumping back to our demo above, here are some alternative commercial APIs… ● IBM Watson Visual Recognition ● Microsoft Computer Vision API ● Clarifai
- 50. ● archive.ics.uci.edu/ml ● deeplearning.net/datasets ● mldata.org ● grouplens.org/datasets ● cs.toronto.edu/~kriz/cifar.html ● cs.cornell.edu/people/pabo/movie-review-data ● yann.lecun.com/exdb/mnist (handwriting) ● kdnuggets.com/datasets/index.html (long list) ● image-net.org (competition) PLUS OPEN SOURCE DATASETS
- 51. ● gallery.cortanaintelligence.com ● bigml.com/gallery/models ● predictionio.incubator.apache.org/gallery/template-gallery ● github.com/tensorflow/models/tree/master/inception ● algorithmia.com/algorithms * AND PRE-BUILT M.L. MODELS
- 52. Start with a pre-trained cloud API (no training data required). Most cloud providers offer a free tier. So start thinking about the different ways to use computer vision. Then just start testing and see what you can do. EASIER THAN YOU THINK R EC A P Don't get me wrong, there's an insane amount of complexity behind the scenes. But fortunately, you need to know that stuff to take advantage of A.I.
- 53. NEXT STEPS
- 54. HOW-TO GUIDES ● Building Voice-Enabled Products With Amazon Alexa ● Cognitive Customer Engagement Using IBM Watson ● Harnessing Visual Data Using Google Cloud ● Building a Recommendation Engine Using Microsoft Azure ● Predicting Marketing Campaign Response Using Amazon Machine Learning ● Unleashing A.I.-Powered Conversation With IBM Watson ● Get into the Mind of Your Customer Using Google’s Sentiment Analysis Tools ● Discover Your Customers’ Deepest Feelings Using Microsoft Facial Recognition ● Give Your Products the Power of Speech Using Amazon Polly ● Computers Are Opening Their Eyes — and They’re Already Better at Seeing Than We Are ● How to Predict When You’re Going to Lose a Subscriber ● The Future of Business is a Digital Spokesperson — Let’s Build a Preview Using Microsoft’s Bot Framework ● Predicting Personality Traits from Content Using IBM Watson R ESO U R C ES How to build the demo app in this session
- 55. ● Computer speech is ready for prime time ● It’s coming faster than you think ● It’s a gateway capability ● It’s changing everything ● It creates a richer experience ● It’s easier to use than you think JOURNEY’S END R EC A P
- 56. COMING UP... Laying the foundation ● Cutting Through the Hype 2 A.I. Technologies that will have the greatest impact ● Computer Speech ● Computer Vision 2 A.I. Applications with the quickest R.O.I. ● Predictive Engagement ● Predictive Personalization STA Y TU N ED
- 57. QUESTIONS OR COMMENTS? Gigaom A.I. Team: ai@gigaom.com Workshop Facilitator: chris.mohritz@10xeffect.com C O N TA C T
- 58. THANK YOU