Trevor Sullivan created GUACAMOLE-1683:
------------------------------------------
Summary: Implement super resolution machine learning model in
Apache Guacamole client
Key: GUACAMOLE-1683
URL: https://issues.apache.org/jira/browse/GUACAMOLE-1683
Project: Guacamole
Issue Type: Wish
Components: RDP, VNC
Environment: Server: Linux virtual machine running on Linode
Client: Windows 11 with Google Chrome web browser
Reporter: Trevor Sullivan
*Summary*
As an end user of Apache Guacamole, I would like improved screen resolution,
especially for text content, without harming network performance. I am using
Starlink satellite internet service, which has highly variable performance,
sometimes fast, sometimes slow and unreliable. Using Apache Guacamole across a
satellite internet connection yields varying results.
I cannot afford to increase screen resolution on the remote VNC Linux server,
because that would severely damage performance, measured in frames per second
(FPS).
*Implementation / Solution*
Implement a super resolution machine learning model into the Apache Guacamole
VNC + RDP web clients. The super resolution model should apply in real-time, to
the video stream coming from the remote server.
*Additional Notes*
I am willing to sacrifice some level of super resolution accuracy, in favor of
improved screen resolution. Machine learning models can be trained to improve
performance to approach closer to 100% accuracy, but will almost certainly
never achieve 100% accuracy.
There is a research paper titled "{_}Implicit Transformer Network for Screen
Content{_}
{_}Image Continuous Super-Resolution{_}" which covers this exact scenario, that
was published in December 2021. Please see this document for details:
[https://arxiv.org/pdf/2112.06174.pdf]
Tensorflow.js allegedly allows developers to implement machine learning models
directly in the web browser. [https://www.tensorflow.org/js]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
