Complexity Performance Evaluation
Because evaluating the RD performance addresses only one side of the problem, in this web page it is also presented an evaluation of the complexity performance for the DISCOVER Wyner-Ziv monoview video codec. Although it is commonly claimed that DVC ‘encoding complexity is low’ and ‘decoding complexity is high’, not much complexity evaluation results are available.
Encoding Complexity
The encoder complexity includes two major components which are the WZ coding and the key frames coding parts. The larger is the GOP size the less key frames are present in the bitstream and thus the lower will be the share of the key frames.
While it is possible to measure the encoding complexity in many ways, some of them rather sophisticated, it is also possible to get a rather good ‘feeling’ of relative complexities using rather simple complexity metrics. In this case, the encoding complexity is measured by means of the encoding time for the full sequence, in seconds, under controlled conditions. It is well known, the encoding (and decoding) time results are highly dependent on the used hardware and software platforms. For the present results, the hardware used was an x86 machine with a dual core Pentium D processor at 3.4 GHz with 2048 MB of RAM. Regarding the software conditions, the results were obtained with a Windows XP operative system, with the C++ code written using version 8.0 of the Visual Studio C++ compiler, with optimizations parameters on, such as the release mode and speed optimizations. Besides the operating system, nothing was running in the machine when gathering the performance results to avoid influencing them. Under these conditions, the results have a relative and comparative value, in this case allowing comparing the DISCOVER codec with alternative solutions, e.g. H.264/AVC, running in the same hardware and software conditions.
The following results were obtained:
Decoding Complexity
The decoder complexity includes two major components which are the WZ decoding and the key frames decoding parts. The larger is the GOP size the less key frames are present in the bitstream and thus the lower will be the complexity share of the key frames. Following the options taken for the encoding complexity evaluation, the decoding complexity evaluation will be measured using an equivalent metric for the decoder, this means the decoding time for the full sequence, in seconds, under the same conditions and software/hardware platform.
The following results were obtained: