Recherche avancée

Sur d’autres sites (8382)

• ffmpeg does not produce result with C# process

  20 août 2020, par Ali Agha

  So, I'm facing a really strange issue. When I run "ffmpeg" through C# process, it doesn't work for multiple commands (concatenated with &). However, it works for simple/single command and even produces output. I checked the logs through "Standard Error" and it appears that it fails in "concat" command.

  


  This line specifically indicates the error in the ffmpeg created log :
concat:CoverImage_Video_Landscape_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.ts|Downloaded_Video_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.ts: No such file or directory

  


  Now, the code snippet is as follows :

  


  string arguments = $"-loop 1 -i \"{physicalPath}bin\\CoverImage_Video_Landscape_{guid}.png\" -f lavfi -i anullsrc -t 3 -c:v libx264 -c:a aac -b:a 160k -t 3 -pix_fmt yuv420p -vf scale={videoWidth}:{videoHeight} -y \"{physicalPath}bin\\CoverImage_Video_Landscape_{guid}.ts\"" 
+ $" & -i \"{physicalPath}bin\\{objectNameWithoutExtension}.mp4\" -c:v libx264 -c:a aac -b:a 160k -f mpegts \"{physicalPath}bin\\{objectNameWithoutExtension}.ts\"" 
+ $" & -i \"concat:CoverImage_Video_Landscape_{guid}.ts|{objectNameWithoutExtension}.ts\" -c copy \"{fileAssetRootPath}\\Video\\Output\\{outputVideoName}.mp4\"";

            process = new Process();
            process.StartInfo.FileName = ffmpegPath;
            process.StartInfo.Arguments = arguments;
            process.StartInfo.CreateNoWindow = true;
            process.StartInfo.UseShellExecute = false;
            process.StartInfo.RedirectStandardOutput = true;
            process.StartInfo.RedirectStandardError = true;

            process.EnableRaisingEvents = true;
            process.Exited += new EventHandler(process_Exited);
            process.Start();
            string stResult = process.StandardOutput.ReadToEnd();
            string erResult = process.StandardError.ReadToEnd();
            process.WaitForExit();


  


  "Physical Path" -> Application directory (bin) folder
"File Asset Path" -> The directory in which I'm storing uploaded resources

  


  Now, the strange part. During debugging, if I just copy the "arguments" variable value and execute it in CMD directly in "bin" folder, I get perfect results. Everything works. Here's the actual query generated :

  


  ffmpeg -loop 1 -i "E:\Projects\Freelance\UP - Inspire3 (Karl)\AWS S3 Bucket Editing Routine\Code\AWSAPI\AWSAPI\bin\CoverImage_Video_Landscape_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.png" -f lavfi -i anullsrc -t 3 -c:v libx264 -c:a aac -b:a 160k -t 3 -pix_fmt yuv420p -vf scale=1920:964 -y "E:\Projects\Freelance\UP - Inspire3 (Karl)\AWS S3 Bucket Editing Routine\Code\AWSAPI\AWSAPI\bin\CoverImage_Video_Landscape_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.ts" & ffmpeg -i "E:\Projects\Freelance\UP - Inspire3 (Karl)\AWS S3 Bucket Editing Routine\Code\AWSAPI\AWSAPI\bin\Downloaded_Video_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.mp4" -c:v libx264 -c:a aac -b:a 160k -f mpegts "E:\Projects\Freelance\UP - Inspire3 (Karl)\AWS S3 Bucket Editing Routine\Code\AWSAPI\AWSAPI\bin\Downloaded_Video_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.ts" & ffmpeg -i "concat:CoverImage_Video_Landscape_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.ts|Downloaded_Video_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.ts" -c copy "E:\Projects\Freelance\UP - Inspire3 (Karl)\AWS S3 Bucket Editing Routine\Code\AWSAPI\AWSAPI\FileAssets\Video\Output\Downloaded_Video_fd8dbbb5-63c0-4ff0-b0c6-e6c3c8f590aa.mp4"


  


  I have added "ffmpeg" keyword before every actual query otherwise I can't run it in CMD. Rest of the query is exactly the same

  


  This query doesn't produce anything when running through C# process and throws the "concat" error I just mentioned above.

  


  FYI, all files needed for conversion are present in the "bin" directory (temp files) and output file is supposed to be placed in another directory. I did that, because concat command apparently has issues with paths, so I just placed ffmpeg.exe along with all files needed directly in "bin" folder.

  


  Also, the code explanation is that it's a 3-step process. It's supposed to convert a png image to ts, then another video to ts, and then combine the 2 and produce output.

  


  Any help would be highly appreciated, thanks !!

  


• Minimal Understanding of VP8′s Forward Transform

  16 novembre 2010, par Multimedia Mike — VP8

  Regarding my toy VP8 encoder, Pengvado mentioned in the comments of my last post, “x264 looks perfect using only i16x16 DC mode. You must be doing something wrong in computing residual or fdct or quantization.” This makes a lot of sense. The encoder generates a series of elements which describe how to reconstruct the original image. Intra block reconstruction takes into consideration the following elements :

  
  
  

  I have already verified that both my encoder and FFmpeg’s VP8 decoder agree precisely on how to reconstruct blocks based on the predictors, coefficients, and quantizers. Thus, if the decoded image still looks crazy, the elements the encoder is generating to describe the image must be wrong.

  So I started studying the forward DCT, which I had cribbed wholesale from the original libvpx 0.9.0 source code. It should be noted that the formal VP8 spec only defines the inverse transform process, not the forward process. I was using a version designated as the “short” version, vs. the “fast” version. Then I looked at the 0.9.5 FDCT. Then I got the idea of comparing the results of each.

  input:   92 91 89 86 91 90 88 86 89 89 89 88 89 87 88 93

  • libvpx 0.9.0 “short” :

    forward : -314 5 1 5 4 5 -2 0 0 1 -1 -1 1 11 -3 -4
    inverse : 92 91 89 86 89 86 91 90 91 90 88 86 88 86 89 89
    

  • libvpx 0.9.0 “fast” :

    forward : -314 4 0 5 4 4 -2 0 0 1 0 -1 1 11 -2 -5
    inverse : 91 91 89 86 88 86 91 90 91 90 88 86 88 86 89 89
    

  • libvpx 0.9.5 “short” :

    forward : -312 7 1 0 1 12 -5 2 2 -3 3 -1 1 0 -2 1
    inverse : 92 91 89 86 91 90 88 86 89 89 89 88 89 87 88 93
    

  I was surprised when I noticed that input[] != idct(fdct(input[])) in some of the above cases. Then I remembered that the aforementioned property isn’t what is meant by a “bit-exact” transform– only that all implementations of the inverse transform are supposed to produce bit-exact output for a given vector of input coefficients.

  Anyway, I tried applying each of these forward transforms. I got slightly differing results, with the latest one I tried (the fdct from libvpx 0.9.5) producing the best results (to my eye). At least the trees look better in the Big Buck Bunny logo image :

  
  
  The dense trees of the Big Buck Bunny logo using one of the libvpx 0.9.0 forward transforms
  
  
  The same segment of the image using the libvpx 0.9.5 forward transform
  

  Then again, it could be that the different numbers generated by the newer forward transform triggered different prediction modes to be chosen. Overall, adapting the newer FDCT did not dramatically improve the encoding quality.

  Working on the intra 4×4 mode encoding is generating some rather more accurate blocks than my intra 16×16 encoder. Pengvado indicated that x264 generates perfectly legible results when forcing the encoder to only use intra 16×16 mode. To be honest, I’m having trouble understanding how that can possibly occur thanks to the Walsh-Hadamard transform (WHT). I think that’s where a lot of the error is creeping in with my intra 16×16 encoder. Then again, FFmpeg implements an inverse WHT function that bears ‘vp8′ in its name. This implies that it’s custom to the algorithm and not exactly shared with H.264.

• Visualizing Call Graphs Using Gephi

  1er septembre 2014, par Multimedia Mike — General

  When I was at university studying computer science, I took a basic chemistry course. During an accompanying lab, the teaching assistant chatted me up and asked about my major. He then said, “Computer science ? Well, that’s just typing stuff, right ?”

  My impulsive retort : “Sure, and chemistry is just about mixing together liquids and coming up with different colored liquids, as seen on the cover of my high school chemistry textbook, right ?”

  
  
  

  In fact, pure computer science has precious little to do with typing (as is joked in CS circles, computer science is about computers in the same way that astronomy is about telescopes). However, people who study computer science often pursue careers as programmers, or to put it in fancier professional language, software engineers.

  So, what’s a software engineer’s job ? Isn’t it just typing ? That’s where I’ve been going with this overly long setup. After thinking about it for long enough, I like to say that a software engineer’s trade is managing complexity.

  A few years ago, I discovered Gephi, an open source tool for graph and data visualization. It looked neat but I didn’t have much use for it at the time. Recently, however, I was trying to get a better handle on a large codebase. I.e., I was trying to manage the project’s complexity. And then I thought of Gephi again.

  Prior Work
  One way to get a grip on a large C codebase is to instrument it for profiling and extract details from the profiler. On Linux systems, this means compiling and linking the code using the -pg flag. After running the executable, there will be a gmon.out file which is post-processed using the gprof command.

  GNU software development tools have a reputation for being rather powerful and flexible, but also extremely raw. This first hit home when I was learning how to use the GNU tool for code coverage — gcov — and the way it outputs very raw data that you need to massage with other tools in order to get really useful intelligence.

  And so it is with gprof output. The output gives you a list of functions sorted by the amount of processing time spent in each. Then it gives you a flattened call tree. This is arranged as “during the profiled executions, function c was called by functions a and b and called functions d, e, and f ; function d was called by function c and called functions g and h”.

  How can this call tree data be represented in a more instructive manner that is easier to navigate ? My first impulse (and I don’t think I’m alone in this) is to convert the gprof call tree into a representation suitable for interpretation by Graphviz. Unfortunately, doing so tends to generate some enormous and unwieldy static images.

  Feeding gprof Data To Gephi
  I learned of Gephi a few years ago and recalled it when I developed an interest in gaining better perspective on a large base of alien C code. To understand what this codebase is doing for a particular use case, instrument it with gprof, gather execution data, and then study the code paths.

  How could I feed the gprof data into Gephi ? Gephi supports numerous graphing formats including an XML-based format named GEXF.

  Thus, the challenge becomes converting gprof output to GEXF.

  Which I did.

  Demonstration
  I have been absent from FFmpeg development for a long time, which is a pity because a lot of interesting development has occurred over the last 2-3 years after a troubling period of stagnation. I know that 2 big video codec developments have been HEVC (next in the line of MPEG codecs) and VP9 (heir to VP8’s throne). FFmpeg implements them both now.

  I decided I wanted to study the code flow of VP9. So I got the latest FFmpeg code from git and built it using the options "--extra-cflags=-pg --extra-ldflags=-pg". Annoyingly, I also needed to specify "--disable-asm" because gcc complains of some register allocation snafus when compiling inline ASM in profiling mode (and this is on x86_64). No matter ; ASM isn’t necessary for understanding overall code flow.

  After compiling, the binary ‘ffmpeg_g’ will have symbols and be instrumented for profiling. I grabbed a sample from this VP9 test vector set and went to work.

  ./ffmpeg_g -i vp90-2-00-quantizer-00.webm -f null /dev/null
  gprof ./ffmpeg_g > vp9decode.txt
  convert-gprof-to-gexf.py vp9decode.txt > /bigdisk/vp9decode.gexf
  

  Gephi loads vp9decode.gexf with no problem. Using Gephi, however, can be a bit challenging if one is not versed in any data exploration jargon. I recommend this Gephi getting starting guide in slide deck form. Here’s what the default graph looks like :

  
  
  

  Not very pretty or helpful. BTW, that beefy arrow running from mid-top to lower-right is the call from decode_coeffs_b -> iwht_iwht_4x4_add_c. There were 18774 from the former to the latter in this execution. Right now, the edge thicknesses correlate to number of calls between the nodes, which I’m not sure is the best representation.

  Following the tutorial slide deck, I at least learned how to enable the node labels (function symbols in this case) and apply a layout algorithm. The tutorial shows the force atlas layout. Here’s what the node neighborhood looks like for probing file type :

  
  
  

  Okay, so that’s not especially surprising– avprobe_input_format3 calls all of the *_probe functions in order to automatically determine input type. Let’s find that decode_coeffs_b function and see what its neighborhood looks like :

  
  
  

  That’s not very useful. Perhaps another algorithm might help. I select the Fruchterman–Reingold algorithm instead and get a slightly more coherent representation of the decoding node neighborhood :

  
  
  

  Further Work
  Obviously, I’m just getting started with this data exploration topic. One thing I would really appreciate in such a tool is the ability to interactively travel the graph since that’s what I’m really hoping to get out of this experiment– watching the code flows.

  Perhaps someone else can find better use cases for visualizing call graph data. Thus, I have published the source code for this tool at Github.