Nvidia RTX 2080 Ti

Total Stations, CAD Workstations, SSD`s, Imaging Systems, Targets & Tripods Etc.
User avatar
smacl
V.I.P Member
V.I.P Member
Posts: 143
Joined: Tue Jan 25, 2011 5:12 pm
Full Name: Shane MacLaughlin
Company Details: Atlas Computers Ltd
Company Position Title: Managing Director
Country: Ireland
Linkedin Profile: Yes
Location: Ireland
Contact:

Re: Nvidia RTX 2080 Ti

Post by smacl » Mon Apr 15, 2019 6:04 pm

dhirota wrote:
Sun Apr 14, 2019 11:32 pm
Shane: I have decided to spend a few US$ to help the economy and expand our research efforts by installing a EVGA Nvida RTX 2080Ti into our i9-9980XE, 18core/36thread, 128GB RAM workstation to test anything that you might produce (hopefully soon) as well as some other software.
Sounds like an absolute beast of a machine Dennis, I'll keep in touch with any up and coming developments. I'm currently at the tail end of a part time course in deep learning for computer vision so have been somewhat swamped but I think it will reap rewards in the longer term. Specifically, if we can align point clouds to HD images (as we do for RGB colorization) and we can use deep learning to identify objects in images, we theoretically ought to be able to use deep learning to classify objects within clouds and then replace those objects with discrete polygon models than get re-colored with the photography. There's quite a bit more to this but it is where I see point cloud processing going and it will certainly be compute intensive! Even some of the old school algorithms for detecting common geometries, such a cylinders for lamp posts really struggle on larger scans and would benefit hugely from being ported to a high-end GPU.

dhirota
V.I.P Member
V.I.P Member
Posts: 519
Joined: Sun Nov 01, 2009 11:18 pm
Full Name: Dennis Hirota
Company Details: Sam O. Hirota Inc.
Company Position Title: President
Country: USA
Linkedin Profile: Yes
Location: Hawaii, USA

Re: Nvidia RTX 2080 Ti

Post by dhirota » Tue Apr 16, 2019 6:46 am

Shane: Your CV course is exactly where you need to be and I am looking forward to see what you learn from it.
smacl wrote:
Mon Apr 15, 2019 6:04 pm
dhirota wrote:
Sun Apr 14, 2019 11:32 pm
Shane: I have decided to spend a few US$ to help the economy and expand our research efforts by installing a EVGA Nvida RTX 2080Ti into our i9-9980XE, 18core/36thread, 128GB RAM workstation to test anything that you might produce (hopefully soon) as well as some other software.
Sounds like an absolute beast of a machine Dennis, I'll keep in touch with any up and coming developments. I'm currently at the tail end of a part time course in deep learning for computer vision so have been somewhat swamped but I think it will reap rewards in the longer term. Specifically, if we can align point clouds to HD images (as we do for RGB colorization) and we can use deep learning to identify objects in images, we theoretically ought to be able to use deep learning to classify objects within clouds and then replace those objects with discrete polygon models than get re-colored with the photography. There's quite a bit more to this but it is where I see point cloud processing going and it will certainly be compute intensive! Even some of the old school algorithms for detecting common geometries, such a cylinders for lamp posts really struggle on larger scans and would benefit hugely from being ported to a high-end GPU.
Two years ago, I posted this thread on the CVPR 2017 Conference in HNL.

viewtopic.php?f=39&t=12054

This was my response to questions from Eugene and Jonathan about recognizing images and objects. At that time I did not have the ability to process information using Ubuntu, but I now have a 36 thread system running Ubuntu and today got my RTX2080Ti running on a 4K monitor under Ubuntu 18.04 and and will be able to use my Intel USB 3.0 Neural Compute Stick enabling rapid prototyping, validation and deployment of Deep Neural Network (DNN) inference applications.
dhirota wrote:
Thu Aug 10, 2017 1:41 am
Eugene and Jonathan

I forgot to mention that the DJI Spark UAV is using the Intel Neural Compute Stick AI chip to track the hand motions to control the UAV by recognizing objects.

"The Neural Compute Stick enables rapid prototyping, validation and deployment of Deep Neural Network (DNN) inference applications at the edge. Its low-power VPU architecture enables an entirely new segment of AI applications that aren't reliant on a connection to the cloud. This allows deep learning developers to profile, tune, and deploy Convolutional Neural Network (CNN) on low-power applications that require real-time inferencing."

You need a x86_64 computer running Ubuntu 16.04 | USB 2.0 Type-A port (Recommend USB 3.0) | 1GB RAM | 4GB free storage space.

Fruit_labelled_1000_563_85.jpg
For those that cannot see the image, it is bunch of fruit identified by contents. Will let you know if I get it running.
You do not have the required permissions to view the files attached to this post.
Dennis Hirota, PhD, PE, LPLS
dennishirota@gmail.com

User avatar
smacl
V.I.P Member
V.I.P Member
Posts: 143
Joined: Tue Jan 25, 2011 5:12 pm
Full Name: Shane MacLaughlin
Company Details: Atlas Computers Ltd
Company Position Title: Managing Director
Country: Ireland
Linkedin Profile: Yes
Location: Ireland
Contact:

Re: Nvidia RTX 2080 Ti

Post by smacl » Tue Apr 16, 2019 8:09 am

dhirota wrote:
Tue Apr 16, 2019 6:46 am
Shane: Your CV course is exactly where you need to be and I am looking forward to see what you learn from it.
Doing it more out of interest in the subject more than anything else if I'm honest. I've been running Atlas for three decades now so the last update to my CV probably talks about the wonders of MSDOS and how we can finally ditch CP/M ;)
Two years ago, I posted this thread on the CVPR 2017 Conference in HNL.

viewtopic.php?f=39&t=12054

This was my response to questions from Eugene and Jonathan about recognizing images and objects. At that time I did not have the ability to process information using Ubuntu, but I now have a 36 thread system running Ubuntu and today got my RTX2080Ti running on a 4K monitor under Ubuntu 18.04 and and will be able to use my Intel USB 3.0 Neural Compute Stick enabling rapid prototyping, validation and deployment of Deep Neural Network (DNN) inference applications.
dhirota wrote:
Thu Aug 10, 2017 1:41 am
Eugene and Jonathan

I forgot to mention that the DJI Spark UAV is using the Intel Neural Compute Stick AI chip to track the hand motions to control the UAV by recognizing objects.

"The Neural Compute Stick enables rapid prototyping, validation and deployment of Deep Neural Network (DNN) inference applications at the edge. Its low-power VPU architecture enables an entirely new segment of AI applications that aren't reliant on a connection to the cloud. This allows deep learning developers to profile, tune, and deploy Convolutional Neural Network (CNN) on low-power applications that require real-time inferencing."

You need a x86_64 computer running Ubuntu 16.04 | USB 2.0 Type-A port (Recommend USB 3.0) | 1GB RAM | 4GB free storage space.

Fruit_labelled_1000_563_85.jpg

Will let you if I get it running.
Great work Dennis, a friend of mine has a DJI spark so I must have a look at it in action. Have you looked at the Google open images library? Lots of training data there. From what I gather, man made objects with low variance don't need nearly as much data as objects with high variance. e.g. there are fewer types of lamp post and pillar than there are types of car, there are fewer types of car than of people. YOLO and Tensorflow both have easy interfaces to get up and running for real time applications, though both come up with bounding boxes rather then bounding polygons and the results can be weak without strong training data. For the data to be properly useful, having identified objects, you then need to estimate pose, which involves having training data with enough distinct key-points. If you go back to the point cloud from the bounding box, you could possibly derive bounding polygons from depth and camera oriented normals and use these in turn to derive pose. You could also do the same using a stereo pair of photographs. Most of what I've seen so far with CNNs and DNNs is based around 2D images as these are the most commonly encountered form of data at this point in time. They're also very explicit, e.g. looking to describe something as a very specific object such as an apple or orange as opposed to a roughly spherical shaped fruit that is orange or red in color, as we see over on Joon's thread. The pose estimation based on key points is actually a very similar problem to cloud to cloud registration, the guys from Correvate did a good demo on this last year.

I suspect that computer vision and AI based object detection today is roughly where laser scanning was when the first Cyrax hit the steets :)

Post Reply

Return to “Other Hardware”