Joseph Redmon: How computers learn to recognize objects instantly
ג'וזף רדמון: כיצד מחשב לומד לזהות אובייקטים באופן מיידי
Joseph Redmon works on the YOLO algorithm, which combines the simple face detection of your phone camera with a cloud-based AI -- in real time. Full bio
Double-click the English transcript below to play the video.
thought that getting a computer
between a cat and a dog
in the state of artificial intelligence.
במצב של הבינה המלאכותית.
greater than 99 percent accuracy.
ברמת דיוק של למעלה מ 99 אחוז.
put a label to that image --
thousands of other categories as well.
at the University of Washington,
באוניברסיטת וושינגטון,
computer vision models.
a prediction of dog or cat,
specific breed predictions.
of granularity we have now.
in image classification,
when we run our classifier
with a pretty similar prediction.
there is a malamute in the image,
we don't actually know that much
איננו ממש יודעים כל כך הרבה
called object detection,
and try to find all of the objects,
ומנסים למצוא את כל האובייקטים,
when we run a detector on this image.
גלאי על התמונה הזאת.
with our computer vision algorithms.
האלגוריתמים של הראייה הממוחשבת.
that there's a cat and a dog.
on top of computer vision,
or a robotic system,
of information that you want.
you can interact with the physical world.
on object detection,
to process a single image.
speed is so important in this domain,
כה חשובה בתחום זה,
to process an image.
it makes predictions,
by another factor of 10,
לפי מקדם נוסף של 10,
at five frames per second.
like this driving my car.
running in real time on my laptop.
בזמן אמת על המחשב הנייד שלי.
as I move around the frame,
כשאני זז סביב המסגרת,
of changes in size,
on top of computer vision.
על גבי ראייה ממוחשבת.
a thousand times faster.
object detection systems
on each of these regions,
על כל אחד מאזורים אלה,
detections in the image.
thousands of times over an image,
אלפי פעמים על תמונה,
to produce detection.
כדי לייצר זיהוי.
to do all of detection for us.
לעשות את כל הזיהוי עבורנו.
and class probabilities simultaneously.
ואת סוג ההסתברויות בו זמנית.
at an image thousands of times
על תמונה אלפי פעמים
the YOLO method of object detection.
שיטת YOLO לזיהוי אובייקט.
we're not just limited to images;
that cat and dog,
את החתול והכלב האלה,
and interact with each other.
ומתקשרים אחד עם השני.
like spoon and fork, bowl,
out into the audience
נוכל לזהות.
our threshold for detection a little bit,
out in the audience.
is happening in real time
object detection system,
already using this technology
שכבר משתמשים בטכנולוגיה זו
like medicine, robotics.
of animals in Nairobi National Park
בפארק הלאומי של ניירובי
of this detection system.
free for anyone to use.
לכל מי שרוצה להשתמש,
even more accessible and usable,
לאפילו יותר נגיש ושמיש,
of model optimization,
של אופטימיזציה של המודל,
running on a phone.
now we have a pretty powerful solution
עכשיו יש לנו פתרון די חזק
and build something with it.
with access to this software,
will build with this technology.
יבנו עם טכנולוגיה זו.
ABOUT THE SPEAKER
Joseph Redmon - Computer scientistJoseph Redmon works on the YOLO algorithm, which combines the simple face detection of your phone camera with a cloud-based AI -- in real time.
Why you should listen
Computer scientist Joseph Redmon is working on the YOLO (You Only Look Once) algorithm, which has a simple goal: to deliver image recognition and object detection at a speed that would seem science-fictional only a few years ago. The algorithm looks like the simple face detection of a camera app but with the level complexity of systems like Google's Deep Mind Cloud Vision, using Convolutional Deep Neural Networks to crunch object detection in realtime. It's the kind of technology that will be embedded on all smartphones in the next few years.
Redmon is also internet-famous for his resume.
Joseph Redmon | Speaker | TED.com