RealSense SDK 2.0 version 2.13.0 released (Update: pre-release status)

[MartyG]

Update: SDK version 2.13.0 has been changed from stable Full Release status to Pre Release status. It is likely to return to a dependable Full release soon.

Original message

********

Hi everyone,

Version 2.13.0 of the RealSense SDK is now available for download. This version replaces 2.12.0, which is no longer available. The major feature added in this version is CUDA optimized implementation for Nvidia Jetson.

https://github.com/IntelRealSense/librealsense/releases/tag/v2.13.0 Release Intel® RealSense™ SDK 2.0 (Build 2.13.0) · IntelRealSense/librealsense · GitHub

Also, it is important to note that the required CMake version to build librealsense project files was promoted to 3.8.

API changes

- rs2_frame_metadata_value - Extending available attributes:

- Depth Sensor:

Laser Power, Laser Power Mode, Exposure Priority, Exposure ROI.

- RGB Sensor:

Brightness, Contrast, Saturation, Sharpness, Backlight_Compensation, Hue, Gamma, White_Balance_Mode & Temperature, Powerline Frequency and Low Light Compensation.

Features

- Additional per-frame metadata attributes available for Depth and RGB sensors.

- Adding CUDA-optimized implementation for Jetson-TX (arm) platform

Bug fixes

- Post-processing filters invalidate metadata

- Enabling D430 in librealsense Demos

- Update kernel patches according to the latest Ubuntu policies

- Unity support for cameras without RGB sensor

- [Realsense Viewer] Save setting fail

- Multi-cam support is broken on some Mac OS systems

******

A list of unresolved Known Issues can be found in the Release Notes.

https://github.com/IntelRealSense/librealsense/wiki/Release-Notes Release Notes · IntelRealSense/librealsense Wiki · GitHub

[RTasa]

Hooray CUDA and Jetson right on time.

[MartyG]

What do you think of the forthcoming Jetson Xavier?

[RTasa]

Sadly again, the Xavier is only game in town.

FAST enough for Deep Learning and setup for robotics/machine learning.

Hope they have a i version for industrial like they have for the TX2.

FYI, I am side tracking here, my largest challenge to date is the Movidius sticks.

I was so excited to use their demos of a clean accelerated inference processor but

you can not install it on a TX2 (at least so far it has failed miserably) nor can you

install it on regular Unbuntu 16.04 laptop (I have older laptop it just fails)

Also.. not being able to just use an inception generated PB file to create a Movidius

graph is dragging me down. So the examples worked OK, but I am having impossible

time making our own stuff work. I am sure they are swamped so I don't complain but I am

not happy. So many challenges this is just adding on. ON the other hand the D435

has been a pleasant surprise now that I have it running.

Love the demo's and if I can get this depth thing figured out I have a direct use for it.

Only thing is do they plan to have an industrial or even automotive grade version?

Current concept is to use it for image segmentation on distance and then send the parts to

be inferenced.

[RTasa]

>>The major feature added in this version is CUDA optimized implementation for Nvidia Jetson.

I was just replied from an intel support person that the Jetson is not supported. Very sad day as I was looking at getting

an Xavier board.

[MartyG]

"Not supported" can sometimes mean "We're not officially supporting that right now", not that it is impossible to do it (as the JetsonHacks site has guides for using the 400 Series with TX2).

Intel's Android and Raspberry Pi 3 Model B guides for the 400 Series shows they are definitely willing to get the camera working with non-PC platforms in some way.

Dorodnic, the RealSense Software Manager, explained in Wednesday's online seminar that Intel are aiming to get SDK 2.0 to work with as many platforms as possible. Sometimes there are difficulties such as a platform that doesn't allow kernel updating, but they are trying to address that with measures such as a bypass of the operating system, and trying to eliminate the need to update kernels at all. Also, they are still learning how to get the SDK to work well with ARM. They're making good progress though.

You can read my notes from the session and the Q&A transcript at the link below.