NihAV — Glue and Hacks

May 20th, 2017

I don’t like to write the code that does nothing, it’s the excitement of my code doing at least something that keeps me writing code. So instead of designing a lot of new interfaces and such that can describe all theoretically feasible stuff plus utility code to handle the things passed through aforementioned interfaces, I’ve just added some barely working stuff, wrote a somewhat working demuxer and made a decoder.

And here it is:
Read the rest of this entry »

#chemicalexperiments

May 19th, 2017

Well, here’s yet another post nobody asked.

As a bog standard programmer I love organ music, hacking various stuff, and cooking Also it’s easier to satisfy my tastes and limitations that way too.

I’m not a skilled cook at all but I can make myself a semi-decent soup or bake something (casserole, quiche or pie). And here’s my short report on trying macaroni and cheese in three variations.

The first version was made after some recipe—cook pasta (I chose fusilli because it’s the only kind I had at hoof), make cheese sauce (essentially start with sauce thickener made from fried flour, add milk and melt a lot of cheese in it), combine together and bake in oven. Simple, filling and tasty. The only problem I found is that it thickens into a solid mass when cooled but it’s still enjoyable then.

The second version I tried was Kraft dinner. Just cook the pasta from the box and mix it with milk, butter and powder (from the packet inside the box) in still warm cooking pot. This version I found incompatible with me—not gross or allergy inducing, just after tasting one spoonful I could not bring myself to take another. Oh well, not a big loss.

And finally, käsespätzle. For this variation you take spätzle (the usual long thin variation sold in every supermarket here), mix it with cooking cream that has been boiled and with some cheese melted it, put the result into baking dish, sprinkle with more grated cheese and bake (I’ve also added chopped dried tomatoes because I had to put them somewhere). The result is tasty and more tender than the first variation. So I approve it too.

P.S. I don’t take pictures of what I cook, you want #opticalexperiments then and from a different person too.

A Short Essay on Bitstream Reading

May 15th, 2017

So, it has come to this. How does bitstream reading might work. Here I’ll try to present several ways to read bits and variable-length codes.
Read the rest of this entry »

Why Modern Video Codecs Suck and Will Keep on Sucking

May 12th, 2017

If you look at the modern video codecs you’ll spot one problem: they get designed for large resolutions and follow one-size-does-not-fit-exactly-anybody approach. By that I mean that codecs are following the model introduced by ITU H.261—split image into blocks, predict block from the previous frame if possible, apply DCT, quantise and code resulting coefficients (using zigzag scan order and special treatment for runs of zeroes). The same was later applied to pictures in JPEG format that is still staying strong.

Of course modern codecs are much more complex that that, current ITU H.EVC standard enhanced every stage:

  • image is no longer split into 8×8 blocks, you have quadtrees coding blocks from 64×64 down to 4×4 pixels;
  • block prediction got more complicated, now you have intra (or spatial prediction) that tries to fill block with gradient derived from already decoded neighbour blocks) and inter prediction (the old prediction from the previous frame);
  • and obviously inter prediction is not that simple either: now it’s decoupled from transformed block and can have completely different sizes (like 16×4 or 24×32), instead of single previous frame you can use two reference frames selected from two separate lists of references and even motion vectors are often predicted using motion vectors from the reference frames (does anybody like implementing those colocated MV prediction modes BTW?);
  • DCT is replaced with some bitexact integer approximations (and the dequantisation and/or transform stages may be skipped completely);
  • there are more scan types used and all values are coded using some context-adaptive coder.

Plus some hacks for low-resolution mode (e.g. special 4×4 transform for luma), lossless (or as they call it, “PCM coding”) and now also special coding mode for screen content (i.e. images with fewer distinct colours and where fine details matter).

The enhancements on streamline coding process are enhancements, they don’t change principles of coding but rather adapt them to modern conditions (meaning that there’s demand in higher compression and there’s more CPU power and RAM can be thrown at the processing—mostly RAM though).

And what the hacks do? They try to deal with the fact that this model works fine for smooth changing continuous tone images and it does not work that good on other types of video source. There are several ways to deal with the problem but keep in mind that the problem of distinguishing video types and selecting proper coding is AI-complete:

  1. JPEG+PNG approach. You select best coder for the source manually and transmit it like that. Obviously it works well in limited scenarios but even people quite often don’t bother and compress everything with the single format even if that hurts quality or compression ratio. Plus you need to handle two different formats, make sure that the receiving end also supports them etc etc.
  2. MPEG-4 approach. You have single format that has various “coding tools” embedded, they can be both full alternative coding features (like WebP has VP8 compression and lossless compression and nothing common between them or MPEG-4 Audio can be coded as conventional AAC, TwinVQ, speech codec or even as a description for synthesised audio) or various enhancement applied to the main coding method (like you have AAC-LC, AAC-Main that enables several features or HE-AACv2 which takes AAC-LC audio and applies SBR and Parametric Stereo to double its channels and frequency range). Actually there are more than forty various MPEG-4 Audio object types (various coding modes) already, do you think there’s any software that supports everything? And looks like modern video codecs head this way too: they introduce various coding tools (like for screen content) and it would be fun to support all possible features in the decoder. Please consider how much effort should be spent on effectively applying all those tools too (and that’s obviously beside the scope of standards).
  3. ZPAQ approach. The terminal AI-complete solution. You are not merely generating bitstream but first you need to transmit bytecode for a program that will decode this bytestream. It’s the ultimate solution—if you can describe the perfect model for the stream then you can compress it the best. Finding an optimal model for given bitstream is left as an exercise for the reader (in TAoCP it would be marked with M60 I guess).

The second thing I find sucky is combinatorial explosion of encoding parameters. Back in the day you had to worry about selecting the best quantisation matrix (or merely a quantiser) and motion vector if you decided to code it as inter-block. Now you have countless ways to split large tile into smaller blocks, many ways to select prediction mode (inter/intra, prediction angle for intra, partitioning, reference frames and motion vectors) and whether to skip transform stage or not and if not whether it’s worth to subdivide block further or not… The result is as good as string theory—you can get a good one if you can guess zillions of parameters right.

It would be nice to have encoder actually splitting video into scene and actors and transmitting just the changes to the objects (actors, scene) instead of blocks. But then you have a problem of coding those descriptions efficiently and even greater problem of automatically classifying the video into such objects (obviously software can do that, that’s why MPEG-4 Synthetic Video is such a great success). Actually it had some use: there was AVS-S standard for coding video specifically from surveillance cameras (why would China need such standard anyway?). In this standard there was special kind of frame for the whole scene and the main share of video was supposed to be just objects moving around the scene. Even if the standard is obsolete its legacy was included into HEVSAVS2 as three or four new special frame types.

Personally I believe that current video formats are being optimised to local minimum, there are probably other coding methods that give larger gain on certain kinds of data, preferably with less tweaking. For example, that was probably the best thing about Daala, its PVQ coding; the rest was nor crazy enough. I have a gut feeling that vector quantisation might be a good base for an alternative approach to building video codecs. And I think it’s better to have different formats oriented for e.g. low-latency broadcasting and video distributing. If you remember, back in the days people actually spent time to decide which segment was coded better with DivX ;-) 3 Fast-Motion or DivX ;-) 3 Low-Motion, so those who care will be able to select proper format. And the rest can keep watching content in VP11/AV2 format. Probably only the last sentence will come to life.

That’s why I don’t expect bright future in video codecs and that’s why my blog is titled like this.

NihAV – io module

May 11th, 2017

I’ve more or less completed nihav::io module, so let’s look at it (or not, it’s your choice).

There are four components there: bytestream reading, bitstream reading, generic integer code reading and codebook support for bitstream reader. Obviously there is no writing functionality there but I don’t need it now and it can be added later when (or if) needed.
Read the rest of this entry »

NihAV Development Progress

May 6th, 2017

After long considerations and much hesitation NihAV finally accepts it’s first developer (that would be me). And for a change it will be written in Rust. First, it’s an interesting language worth learning; second, it seems to offer needed functionality without much hassle; third, it offers new features that are tempting to try. By new features I mostly mean enums, traits and functions bound to structures.

Previously I expressed the intent to do a completely new design of multimedia (mostly decoding) framework with decoders being assembled from smaller blocks. For example, if I’d implement VIVO H.263 decoder (just as a troll) it would contain these bits:

  • generic 8×8 block decoder interface that does common stuff for such decoders (maintaining block indices, filling frame information e.g. block type, motion vectors etc etc);
  • trait for 8×8 block decoder implementation that does actual bitstream decoding (functions for decoding GOP/picture/slice headers, MV prediction, block data decoding and such);
  • IPB frame shuffler as implementation of generic frame shuffler (i.e. that piece of code that selects which frames to use as references and which frame to output after decoding the current one);
  • maybe even custom codebook accessor (the piece of code that tells codebook generator what is the code and symbol at position N) so it doesn’t need to be converted into some fixed form.

There’s not much code written yet but there are some bits implemented: rudimentary universal bitstream reader, bytestream reader (the same for memory and file I/O) and semi-working framework for demuxing. That’s a start at least.

Side rant: it seems that visible languages (i.e. not completely obscure ones) that use := form assignment have rather unpleasant evangelists (full of themselves in the best case, actively pushing their language to replace everything else in the worst case). That includes Go, Oberon and Wolfram Language. I don’t mean that other languages are free of that problem but in these cases looks like the majority of posts or articles about such language are written from this position.

Blog Restarted

May 6th, 2017

As you (imaginary person who actually read this blog) I’ve stopped blogging last year. Mostly I did it because I ran out of material to write about. Well, now I have some new thoughts that I’d rather dump on the blog and forget so the blog restarted.

While the previous blog was mostly centred on codec reverse engineering and design and rants about random topics, this version should be more about NihAV development and codecs design. And obviously useless rants on random topics—except on opensource multimedia projects (I’ve said enough about CEmpeg, Libav and such; corporate ones I might still have something about).

Let’s see how well it goes this time…

Telegony in FLOSS

July 5th, 2016

In case you don’t know telegony is a belief that all offspring of a female will look like her first partner (there’s nothing wrong with it unless you know biology). In the metaphorical sense it means that the original circumstances of the company or project founding (i.e. who, with what goal and such) will affect its development to the end. Like there was a small company producing so-called “ever sharp” pencils that’s still afloat after a century—but who cares about pencils from that Sharp Corporation nowadays? Anyway, let’s talk about various opensource multimedia projects and see how it applies to them.

FFmpeg. The project was created by FFabrice Bellard who’s known for his brilliant software with horrible source code (he’s won IOCCC twice after all). FFmpeg obviously follows the suit and still has a lot of horrible code that people can’t clean up to this day. And yet it’s as ubiquitous as LZEXE packer back in DOS days or even more.

MPlayer. The project was originally created because its author could not find a video player that would support MPEG-2 video and AVI codecs at the same time—so he hacked something from libmpeg2 and avifile. And what do you know, after all these years it strives to play everything by means of ugly hacks. Those include but not limited to: catching SIGSEGV during MPEG-2 video decoding and restarting decoder, patching loaded VfW/DShow/DMO decoders based on .dll name, calling private internals of bundled FFmpeg.

MPlayer forks. mplayer2 was created by Uoti Urpala who had his disagreements with the original MPlayer team and made his own version with streamlined build system and throwing out many hacks of the original code. mpv seems to follow the suit and I guess if the next fork appears it’ll be also done by a disgruntled developer who wants to throw more hacks out and make it more average-user-friendly.

libvpx. Originally (VP3 times) it was a codec using rather well known coding methods, definitely not state of the art, with a confusing source code that was opensourced without any format description beside source code because the company didn’t care about it but wanted it to become multimedia standard anyway. The company name was On2 then, mind you! But if you look at VP8 or VP9 nothing has changed much (I still have the impression that VP9 format specification was written by engineers from that company implementing hardware decoder and benevolently edited and released by Baidu when they don’t care about VP9 any more because there’s VP X coming).

Derek B. A brilliant reverse engineer who did his first codec (VBLE) by struggling with disassembly and then asking the codec author for the sources. I’ve mentioned before that his best decoders were REd in a very original way: by declaring the intent to do that and waiting for somebody else to do it. Obviously not a project but his story fits so I thought it’s appropriate to mention him here.

VLC. As you might remember the project started to justify a high-speed LAN in some French École Abnormale (because it was not the only high school in France named École Normale in case you wonder). And what do you know—the project went fine as a business project and does so till this day. Though if they ever switch from Discworld-themed naming to Ubuntu-like they should go with Glorified Gstreamer. Why? Because while venerable old projects like XAnim, Xine or MPlayer demonstrated high-class by having their own codec reverse engineering work, VideoLAN went the way similar to the previous story—waiting until somebody does it elsewhere and then announcing support for that format.

Moral of the story: think before you start a project, people may make fun of it afterwards.

A Rant on Actimagine VX Codec

July 2nd, 2016

Well, since I don’t do anything useful these days and just give rants on subjects nobody cares about here’s another one.

This codec (there’s also VX container with IIRC PCM audio to accompany video) can be named a Very Mobile H.264 Rip-off. Why? Because the only available binary specification I got seems to come from some game, in ARMv6 I guess and has stride hardcoded to 256 pixels which would be appropriate only for some hand-held consoles. As for the second part—it uses Elias Gamma’ codes (like H.264 does—under different name) and suspiciously similar spatial prediction. Obviously it also differs a lot because it’s intended for a low-power devices and low resolutions too.

So, it operates on 16×16 macroblocks, each one can be coded with one of 24 possible modes that are really just combinations of one of the following techniques with optional residue coding:

  • splitting into 16×8 or 8×16 sub-blocks and processing them in the same way;
  • copying data from one of the previous three frames with or without motion vector adjustment (it’s full-pixel only);
  • copying data from the previous (frame?) block with some offset added to it (actually three offsets—one per component) and motion vector optionally;
  • applying intra prediction that also comes in two flavours—four modes applied to any block size or nine modes applied for 4×4 luma blocks (still only four modes for chroma).

Residual 16×16 block is coded as 5-bit CBP (again, Elias Gamma’ code mapped to CBP value) for four 4×4 luma blocks and two 4×4 chroma blocks. Coefficients are coded like this:

  1. Mode from Xine table (it’s predicted from neighbouring blocks too) that defines how many coded coefficients are there and how many of them are ones;
  2. Signs for known ones;
  3. Elias Gamma’ codes for other coefficients and their signs;
  4. Zero run value for skips between elements (from Xine table depending on maximum coefficient level seen so far).

In other words—nothing like H.264 CAVLC mode at all.

And if you think it was fun to RE I can tell you it was not and there are still challenges to overcome. First, the specification is badly written and optimised too much that decompiler is almost worthless there (for example, refilling bits is done by jumping to the end of the function that reads unsigned Elias Gamma’ code), functions expecting certain registers to be used for the state (like block functions expect R11 and R12 to contain motion vector, bitstream reading functions operate on the context stored in R1-R3 and return result in R6 etc etc), Hex-Rays also can’t decompile anything with switch statements and block decoding functions are full of them, it often decompiles function just to the first function call and ignores the rest of function code (happened to me on x86 too once where it decided to decompile only the head of main Smacker video decompression function without the block decoding loop, that’s why I trust decompilers even less that compilers). Second, the specification seems to miss data for lookup tables used in coefficients decoding.

So if you want to have it fully REd find a better specification and/or more patient and persistent man to deal with it. As for me—it’s dead, Jim®.

On FFmpeg and Voting

June 27th, 2016

Sometimes I hear that voting in FFmpeg is worse than in North Korea. Obviously people don’t know much about voting in North Korea or they wouldn’t make such statements. Here I shall try to have a short overview on voting in two similar yet different Asian countries and compare it to FFmpeg. My knowledge about North Korea comes from posts of Russian researchers, at least one of them lives in Korea and their information looks legit to me (because it lacks political agenda and has even small details mentioned that correlate well with other information from people visiting DPRK). For the other country it’s various sources not from state media (because such information is often omitted there or you hear simply outright lies). For FFmpeg I obviously have my own experience and observations from their mailing list.

North Korea

Maybe the main reason why it makes people think about FFmpeg is the fact they claim sovereignty over the whole Korean peninsula and even appoints special people to do merges govern over regions still under occupation and even promotes them time from time. South Korea actually does the same except for useless staff but they recognize people with North Korean passports as their own citizens.

Anyway, voting. The system is rather simple: you have one candidate per region and all people vote for him. The ruling member of Kim family should get 100% support in his region (around Paektu Mountain if you forgot that)—and 100% means that all voters should come and vote (I guess it’s obvious how). In other regions some voters may skip it in case they are very ill. I don’t remember whether they can vote against but it’s rather unthinkable too. There’s a story about North Koreans fleeing to China and seeing a demonstration in South Korea on the TV—their reaction was “Why do they allow it?! How can the President run such a country?!”. So people there are quite well conditioned and voting goes smooth.

Russia

This is a country with a variety of approaches to voting—some elections nobody cares about and they can be even fair (until an inappropriate candidate is elected and then they have to correct it), some elections try to keep an image of honesty in order to show outsiders that the system works fine and elected candidates are legitimate (in that case they simply try not to rig it as blatantly), some elections are rigged in the most blatant way and even more and in the ideal case there are no more elections (under some stupid pretext they’ve got rid of governor elections and now in some towns and cities mayors are not elected either because that saves money). And referendums there while still theoretically possible can’t be on any question related to government or status of some region or the questions that are decided by government.

So depending on luck in some elections a vote can matter, in other cases it doesn’t matter and in some cases it matters that much that the votes are cast and counted without voter’s participation (some regions constantly report that over 100% of voters were present and there’s a Russian meme 146% that appeared because when they reported result for parliament elections in Rostov region IIRC the percents from different parties added up to that number and there were some other regions with total sum over 100%).

There are many tricks to get desired results—inventing new demands to filter out unwanted candidates (their neighbour Belarus has a joke “In order to be eligible as presidential candidate the person must have at least five years of presidential experience”), the same voters voting again and again (because there’s a special document allowing a voter to vote in other region—and groups of people can have a dozen of such permits per member and thus vote repeatedly in several places), putting a sheaf of “properly” filled ballots into ballot box or simply counting them as you see fit no matter what was the actual vote there.

Similar story with petitions—they are often masked with similar petitions or later an expert group finds that petition to be infeasible or contradicting the law.

FFmpeg

In the old days voting was usually called mostly on naming issues and it worked. But then disagreement with the practices of the FFmpeg leader (like committing code without any review and despite objections) escalated to the point of The Split but before that there were ugly votings that included old MPlayer members that nobody cared about in FFmpeg (because the leader said everybody with commit rights on svn.mplayerhq.hu should have a vote). So after The Split and another ugly voting there were two projects. I don’t remember any voting in Libav but in FFmpeg this tradition still holds. Like recently there was a voting committee formed and there were at least two serious votes (not just a new season logo in trac)—for code of conduct (because every project should have one but not necessarily follow it) and for banning Carl Eugen Hoyos for some time. The first one obviously passed, the second one rather expectedly failed. But then A Person Known For Resigning As Leader took an action that should be allowed only to leaders and banned some people for 24 hours from the mailing list.

Well, I think it’s clear now that FFmpeg voting is not on North Korea level because there’s animosity there that can be expected only from current MPlayer team but not FFmpeg. But is FFmpeg on par with Russia? Maybe not yet but it tries hard IMO.