[ZJUISLAB]Several questions about video encoding on Tegra K1


Here are several questions about video encoding on tegra k1

  1. The tegra k1 data sheet shows that Multi Standard Video Encoding including MVC is supported on TK1. But we can’t find any APIs or interface to perform MVC (multi view encoding). We have tried “gstreamer” framework on the Jetson TK1 platform, and no MVC related information was found.
    So the question is :
    Is MVC supported in L4T on the Jetson TK1 platform? Is there any APIs or libs available ? If MVC is supported, how many params we can use? And how is the performance and quality?

  2. We want to know the feasibility of developing h.264 high profile encoder(30fps) on the CUDA platform(on TK1). We have known that there are two offical video encoders(NVCUENC and NVENC). NVENC is the lastest version which is Hardware-Accelerated-Based encoder. But only the software-based(CUDA) encoder has a certain reference value for us. We want to know if the older version NVCUENC is only based on CUDA software.
    So the question is:
    Did NVCUENC use any hardware resources which we can’t access? Is it possible to develop h.264 high profile encoder just on the CUDA platform(on Tegra K1) like NVCUENC?

  3. We also want to use gstreamer to perform h.264 high profile encoding.
    refer to the post https://devtalk.nvidia.com/default/topic/851881/embedded-systems/-jetson-tk1-h-264-encoder-high-profile/
    Any way, after replace the “libnvomx.so” by a new one(offerd by FAE), high profile is supported. But CABAC is not supported yet. We want to know is it possible to open CABAC option?

ZJU islab

This thread might have some of the answers you’re looking for, at least to the first question: https://devtalk.nvidia.com/default/topic/778280/embedded-systems/jetson-goes-curling-or-simultaneously-viewing-multiple-ip-cams-/1/