I'm looking for a reference telling me how much the encoding process is more complex than the decoding process in H.264/AVC to use it in my thesis.
I remember a number around 10 times in which the encoder is more complex than the decoder, but I need a reliable reference to make me sure.
Thanks in advance.
I suggest to get in touch with Heiner Kirchhoffer. He's also on ResearchGate and works as researcher at the HHI.
The encoder complexity is highly depends on profile and settings. For example, for x264 fast software implementation the encoder complexity in veryslow preset is around 30 times higher than in ultrafast preset. So, the answer depends on the application (real-time encoding/decoding, off-line encoding and real-time decoding and so on).
As Evgeny has mentioned, encoder complexity can vary substantially based on the tools you wish to support, the intelligence of how these tools will be supported, the target application and so on. An encoder can be quite flexible and can be as simple as possible (i.e. supporting only intra encoding with a fixed number of modes, simple mode decision etc) or as complex as possible (multiple coding passes, a large search range using FS (or a fast search), all possible coding modes (with or without a fast mode decsion), use or not of weighted prediction, use of rate distortion optimization (and the definition of what "distortion" is can vary, and so on. The profiles you may wish to support and the formats you are trying to code (e.g. is it 4:2:0 8 bit only or do you desire to go even to 4:4:4 12bit encoding? How about interlace and the tools supported there and so on) and how you wish to deal with those would also have significant implications, whereas the target platform that you will use for your implementation (software or hardware encoder, can some things be done in parallel, how can I reuse data and so on), also impact even he definition of an encoder's complexity substantially. Performance (as in for example "Quality" given a particular rate target), in the end, is a rather defining term of what the complexity of the encoder would be.
On the other hand a decoder's complexity is much more straightforward. This of course is assuming you wish to be compliant to a particular profile and level as those are defined in the AVC/H.264 specification, and you are also ignoring concepts such as "error resilience" (which interestingly enough most actual decoders do). Although there could be a variety of ways that you can do the same thing, i.e. decode the stream as the specification tells you too, you always need to do that exact thing, otherwise you wouldn't be conforming or will not be able to properly decode your stream. Use of SIMD and multi-threading could of course result in some decoders being faster than others, as well as how data are organized and used during decoding, but that clearly depends on how intelligent is your implementation and not how you utilize the standard.
In any case, if you wish to add some references, maybe checking the early CSVT papers on AVC would be the best for you. Please check Issue 7, Vol.13 in particular (http://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=27384) where you can find lots of related papers such as "H.264/AVC baseline profile decoder complexity" by M. Horowitz et al, and "Complexity of optimized H.26L video decoder implementation
" by V. Lappalainen at al. Obviously reading and referencing the overview paper by the standard's co-chairs is strongly recommended.
@Stefan, @Evgeny and @Alexis. Thanks a lot, for your useful information. I'm checking out the papers you mentioned, they're are exactly what I was looking for.
This is so valuable information for me. As I am right now working on SIMD implementation of x265..Thank U all...
- Badges
- Science topic
- Similar topics
- Correction