My paper of choice was "Real-Time Network Streaming of Dynamic 3D Content with In-Frame and Inter-Frame Compression", by Paolo Simone Gasparello, Giuseppe Marino, Filippo Banno, Franco Tecchia and Massimo Bergamasco.This article gives an insight into, what technical solutions should we consider while implementing the backbone of our service to achieve maximum client satisfaction.
The problem of keeping up with a real-time streaming demand is not new. The article refers to prior works, where developers already encountered and tried to solve the problem of the rendering lag and network restriction hurdles. One of the solution is, to compress, distribute, then serve a request-response pair. The article talks about how 3D objects can be processed and rendered via different methods and at the same time, keeping the process invisible to the clients.
One of the interesting aspects of the paper, that they actually utilized and OpenGL stream distribution methodology, where they intercept API calls and distribute them throughout the dedicated slave nodes. On top of the distributed technology, each frame that is being sent to the backbone network is compressed, to save bandwidth and increase response time. Even on top of this, the authors proposed a method, to not send the whole frame every time; they rather approached the problem by only sending and processing the difference between two frames, further reducing computing and bandwidth requirements. While and OpenGL stream is not too much of a heavy-weight protocol, the biggest problem is to transport a full 3D model, since it can be very complex, thus eating up a lot of bandwidth. To address some of these challenges, they implemented two additional layers between the application and the network, where those layers are responsible for data transmission synchronization (since it is UDP based transmission) and data compression tasks.
The authors talk about frame-to-frame coherence, which refers to, that the actual model is only being "explored" from different angles. By knowing the surroundings and the properties of the models available for us, we can just make changes in the camera position, which simulates the effect, if we would have gotten a new image in every single frame. They also mention, that different texture compression methods are being used, and their implementation is capable to work with different solutions as well.
During testing, they concluded, that the different optimization and compression approaches that they implemented worked well by complementing each other. There is a lot of ideas and future work that can be implemented along this line of research.
Thanks for reading
--
Balazs Makany
No comments:
Post a Comment