Recherche avancée

Sur d’autres sites (5278)

• Out of memory on ffmpeg when converting to H265

  18 août 2017, par Lion

  I’ve a bunch of video files, mostly H264. To save storage, I wrote a batch script, that converts all of them to H265 using ffmpeg. Problem : Some files cause ffmpeg to use ALL my memory (24 GB). Then it crashes (cause it try to allocate even more RAM), which stops the converting process.

  I think that these files are corrupt in some kind. Because with other files, it works well with low memory consumption. Now I want to reject those broken ones, so that unattended converting is possible.

  How is it possible to detect such corruption ? Can ffmpeg do this, or is a third party tool required ?

  My ffmpeg call

  set crf=20
  
  set codec=265
  
  
  
  ffmpeg -hide_banner -i "!fullSourcePath!" -c:v libx%codec% -crf %crf% "%targetPath%\!targetFileName!"

  mkvalidator can’t help

  mkvalidator says that a corrupt file is valid :

  mkvalidator.exe "V:\Filme\_LegacyFormat\22 Jump Street.mkv"
  
  ........................................................................................................................
  
  WRN0D0: There are 5306 bytes of void data..
  
  
  
  mkvalidator 0.5.0: the file appears to be valid
  
          file created with libebml v1.3.0 + libmatroska v1.4.1 / mkvmerge v6.9.1 ('Blue Panther') 64bit built on Apr 18 2014 18:23:38

  eac3to331 can’t help, too

  I found the tool eac3to331, which has a check flag. But it gave me no errors, although the tested file seems corrupt (cause my PC to crash after several minutes running ffmpeg)

  eac3to.exe -check "V:\Filme\_LegacyFormat\22 Jump Street.mkv"
  
  MKV, 1 video track, 2 audio tracks, 1 subtitle track, 1:51:57, 24p /1.001
  
  1: h264/AVC, English, 1920x808 24p /1.001 (240:101)
  
  2: DTS, German, 5.1 channels, 1509kbps, 48kHz
  
  3: DTS, English, 5.1 channels, 1509kbps, 48kHz
  
  4: Subtitle (SRT), German
  
  v01 Extracting video track number 1...
  
  a02 Extracting audio track number 2...
  
  a03 Extracting audio track number 3...
  
  s04 Extracting subtitle track number 4...
  
  Video track 1 contains 161039 frames.
  
  eac3to processing took 1 minute, 26 seconds.
  
  Done.

• Is avcodec_send_frame thread safe for a single AVFrame ?

  22 juin 2024, par Александр А

  I'm doing video transcoding live stream. The H264 input stream from the usb camera, I decided to encode the same AVFrame in different threads for streaming x264 with low bitrate and for jpg snapshots, but at some points avcodec_send_frame is successful, at others it returns error -22, and sometimes even Segmentation fault.

  


  What can be done about it ?

  


  My snapshot handler is for example

  


  int SnapshotEncode(int (*callback)(const char * topic, const uint8_t * buf, int lbuf), const char * topic)
{
    const AVCodec * pCdc = avcodec_find_encoder(AV_CODEC_ID_MJPEG);
    if (!pCdc)
    {
        fprintf(stderr, "ERROR find snapshot encoder for '%s'\n", avcodec_get_name(AV_CODEC_ID_MJPEG));
        return -1;
    }

    struct stEncoder_t * enc = EncoderNew("snapshot");
    if (!enc) return -2;

    int rt = 0;
    AVCodecContext * pCdcCtx = avcodec_alloc_context3(pCdc);
    if (!pCdcCtx)
    {
        PrintErrorAV("alloc snapshot an encoding context", (rt = AVERROR(ENOMEM)));
        return rt;
    }

    pCdcCtx->time_base = DEFAULT_TIME_BASE;
    pCdcCtx->pix_fmt = pCdcCtxInp->pix_fmt;
    pCdcCtx->height = pCdcCtxInp->height;
    pCdcCtx->width = pCdcCtxInp->width;

    if ((rt = avcodec_open2(pCdcCtx, pCdc, NULL)))
    {
        PrintErrorAV("open snapshot encoder", rt);
        goto END;
    }

    AVPacket * pPkt = av_packet_alloc();
    if (!pPkt)
    {
        PrintErrorAV("allocate snapshot AVPacket", (rt = AVERROR(ENOMEM)));
        goto END;
    }

    EncoderRun(enc, 1);
    if ((rt = sem_wait(&enc->sem)) < 0)
        fprintf(stderr, "ERROR lock encoder semaphore: %d %s\n", errno, strerror(errno));

    if (rt)
    {
        EncoderRun(enc, 0);
        pthread_barrier_wait(&stDec.bar);
        goto END;
    }

    if ((rt = avcodec_send_frame(pCdcCtx, stEnc.pFrm)))
        PrintErrorAV("snapshot encode send", rt);
    else if ((rt = avcodec_receive_packet(pCdcCtx, pPkt)))
        PrintErrorAV("snapshot encode receive", rt);

    EncoderRun(enc, 0);
    pthread_barrier_wait(&stDec.bar);

    if (rt == 0) callback(topic, pPkt->data, pPkt->size);

    av_packet_unref(pPkt);

END:
    EncoderFree(enc);
    if (pPkt) av_packet_free(&pPkt);
    if (pCdcCtx) avcodec_free_context(&pCdcCtx);

    return rt;
}



  


• Paperclip geometry ignored

  27 mars 2017, par ACIDSTEALTH

  I have a model called Snapshot, which represents a user-recorded video. I want to take the input video file and scale it to fit within a 720x720 box (ImageMagick documentation). I then want to capture some screenshots of the video to represent it in my app. These sizes are specified accordingly in my model.

  I expect the output of this to be an original video with a maximum width of 720px (assuming it was recorded in landscape mode), a large JPG image with a maximum width of 540px, etc.

  When I attach the video file and save the model, the images and video file are processed but the result is not what I expected. The video file has a resolution of 960x720 and the images are all square (540x540, 360x360, etc).

  I’m not sure if I’m doing something wrong or if this is just a bug with Paperclip. Here is my code :

  class Snapshot < ApplicationRecord
  
      has_attached_file :video,
  
          styles: {
  
              original: { geometry: "720x720", format: 'mp4' },
  
              large: { geometry: "540x540", format: 'jpg' },
  
              medium: { geometry: "360x360", format: 'jpg' },
  
              thumb: { geometry: "180x180", format: 'jpg' }
  
          },
  
          default_url: "", processors: [:transcoder]
  
      validates_attachment_content_type :video, content_type: /\Avideo\/.*\Z/
  
      validates_attachment_size :video, less_than: 100.megabytes
  
  end

  I have also tried adjusting the geometry to 720x720>, 540x540>, etc. When I did this, the attachments’ resolution was unchanged, which seems to go completely against my understanding of how ImageMagick geometry works. I have numerous other models with image-only attachments that do not suffer from this issue.

  Here is a snippet from my Gemfile.lock so you can see which versions I am running :

  delayed_paperclip (3.0.1)
  
    activejob (>= 4.2)
  
    paperclip (>= 3.3)
  
  paperclip (5.1.0)
  
    activemodel (>= 4.2.0)
  
    activesupport (>= 4.2.0)
  
    cocaine (~> 0.5.5)
  
    mime-types
  
    mimemagic (~> 0.3.0)
  
  paperclip-av-transcoder (0.6.4)
  
    av (~> 0.9.0)
  
    paperclip (>= 2.5.2)
  
  paperclip-optimizer (2.0.0)
  
    image_optim (~> 0.19)
  
    paperclip (>= 3.4)

  Update
  I retried this with a video recorded in portrait mode (iPhone 6 front-facing webcam) and the video file’s output size was 720x720. The images were still square as before.