# AI Learns Human Pose Estimation From Videos | Two Minute Papers #237

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=dxOHmvTaCN4
- **Дата:** 17.03.2018
- **Длительность:** 2:55
- **Просмотры:** 32,195
- **Источник:** https://ekstraktznaniy.ru/video/14495

## Описание

The paper "DensePose: Dense Human Pose Estimation In The Wild" is available here:
https://arxiv.org/abs/1802.00434
http://densepose.org/

Our Patreon page: https://www.patreon.com/TwoMinutePapers

We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Andrew Melnychuk, Brian Gilman, Christian Ahlin, Christoph Jadanowski, Dennis Abts, Emmanuel, Eric Haddad, Esa Turkulainen, Evan Breznyik, Frank Goertzen, Malek Cellier, Marten Rauschenberg, Michael Albrecht, Michael Jensen, Nader Shakerin, Raul Araújo da Silva, Robin Graham, Shawn Azman, Steef, Steve Messina, Sunil Kim, Torsten Reil.
https://www.patreon.com/TwoMinutePapers

One-time payment links are available below. Thank you very much for your generous support!
PayPal: https://www.paypal.me/TwoMinutePapers
Bitcoin: 13hhmJnLEzwXgmgJN7RB6bWVdT7WkrFAHh
Ethereum: 0x002BB163DfE89B7aD0712846F1a1E53ba6136b5A
LTC: LM8AUh5bGcNgzq6HaV1jeaJrFvmKxxgiXg

Music: Antarctica by Audionautix is licensed under a Creat

## Транскрипт

### Segment 1 (00:00 - 02:00) []

dr fellow scanners dit is toen meneer pepper sweet caruso knife share this project is een collaboration between in lea en facebook een jaar research and is about proces team is er boos als team is er mee eens dat we teken input foto of in de koeler kees video of people and the output should be a description of their posters this is kind of like motion capture foto's en meezing moevie en computergame animations baas hoe laat de studio en de markers dit workflows ivens further and tries to offer for free the reconstruction of de geometry of de buddy's en het is en vaak doing re more than that enzym ace in hem en het bureau networks and usually great app die tests voorwaarde dat we have large number of trainingsaanbod to train dan zo de first up is gathering large amount of notk this means een input fotograaf of someone who share dag goede correct description of daar past er this is what we call my trainingsaanbod this new propose deden set contains 15. 000 of this training samples and using that we can proceed to stap number to group training de neuraal netwerk de perform pols estimation baat deze more to this particular work normally disposed emission takes place voor de to the skeleton moet smeren stad moest acne is output de stick video met not in this case week als de details uit kunt een segmentation and des correspondances between two the images and 3d models therefore de network is ofzo label de output fully familie marvel's daar plenty of interesting die telefoon in de paper for instance sinds die ene type grande rue foods in de trainingsset is created by humans with plenty of missing dhr is fiat in by using seperate nu al network de specialized for this task make sure to have look at the paper for more cool details like this assange bed in serie met de practical application has to be a robust gaston cushions and dropping changes in posture and the good thing is dat die offers habilis plenty of ik samples we die is dat je kassier absoluut has to be able to deal with smaller en bieder ski-jas met bi-polar closer of further away from de camera dit is absoluut challenge the ugly john darley de job at snv member no markers voor studio setup is required and en richting dat u zich hier is vervormd interactively de details edwina pierson end it will be possible to use it for future research words suspect plenty of molecuul operation and follow-up words for this problem while leaving en meezingt eens in detail searching for your generous support na see you next time [Muziek]