C'est possible de récupérer la présentation ? Et savoir quand auront lieu
les événements.
New office features via Google Drive
Today Google introduce some new office Quickoffice features. Display some awesome kids who made an awesome app, and talked about a lot of changes in ...
Google I/O 2013 - Cloud Platform Track Kickoff: Ushering in the Next Generation of Cloud Computing
Greg Demichillie, Urs Hölzle Senior Vice President Urs Hölzle will share Google'????s vision for the next generation of cloud computing. He'll discuss how ...
Google's Project Tango depth-sensing tablet dev kit (//liliputing.com.)
Google I/O 2012 - Urs Hoelzle's Keynote on Google Cloud Platform
A long time ago at Google, we started working very hard to build the world's fastest, most scalable, and most reliable infrastructure. During the Google I/O 2012 ...
where the list of or any such apps that integrate with this capability?
Google Project Tango - Real Time Localization under Viewpoint Changes
A video showing real time localization where the viewpoint between the mapping run and the localization run showed moderate viewpoint differences. Given the ...
A video showing real time localization where the viewpoint between the
mapping run and the localization showed moderate viewpoint differences.
Given the low-cost interest-point detector and descriptor types used,
achieving good recall despite the changes in viewpoint remains a challenge.
In the right video the previously captured map is overlaid, which shows
that the alignment of camera and map is performed in real-time.
Here we optimize for runtime s.t. we can deliver a 6-DoF pose "cold
starting" (from the raw image without any prior information) within 11ms
(Intel i7) / 65ms (NVidia Tegra K1). The recall achieved is still not
perfect reaching a mere 89% @ 100% precision. Most of the missed frames are
however single frames, which are trivial to bridge by odometry.
I agree with the comment you make on missed frames, however I also argue that all points back from Tango, or any other similar system have to be assumed bad - i.e. they are individually inaccurate and statistically accurate - using point clouds alone leads to images that look OK, but the data has serious problems - I would love to see this example rerun with the sample points voxelized to show what's really there and consistent - I do not mean to slight the work, I have a deep appreciation for the effort - I am only observing that point clouds that look good are still bad :-( However, we all know that's a question of time, not raw capability :-) As far as the other post about pose estimation, drooling I am :-)