Recherche avancée

Sur d’autres sites (7246)

• utvideoenc : Enable support for multiple slices and use them

  14 février 2014, par Jan Ekström

  utvideoenc : Enable support for multiple slices and use them
  The official Ut Video decoder only threads with slices, thus until
  
  now any files encoded by the libavcodec encoder have only been
  
  decodable with a single thread. The default slice count is now
  
  set to subsampled_height / 120.
  Also sets slices to 1 for the Ut Video encoder tests to keep them
  
  green.
  Signed-off-by : Derek Buitenhuis <derek.buitenhuis@gmail.com>
  

  • [DBH] libavcodec/utvideoenc.c
  • [DBH] tests/fate/utvideo.mak

• Issue in recording video

  16 novembre 2015, par human123

  I am trying to record video in 480*480 resolution like in vine using javacv. As a starting point I used the sample provided in https://github.com/bytedeco/javacv/blob/master/samples/RecordActivity.java Video is getting recorded (but not in the desired resolution) and saved.

  But the issue is that 480*480 resolution is not supported natively in android. So some pre processing needs to be done to get the video in desired resolution.

  So once I was able to record video using code sample provided by javacv, next challenge was on how to pre process the video. On research it was found that efficient cropping is possible when final image width required is same as recorded image width. Such a solution was provided in the SO question,Recording video on Android using JavaCV (Updated 2014 02 17). I changed onPreviewFrame method as suggested in that answer.

      @Override
  
      public void onPreviewFrame(byte[] data, Camera camera) {
  
          if (audioRecord == null || audioRecord.getRecordingState() != AudioRecord.RECORDSTATE_RECORDING) {
  
              startTime = System.currentTimeMillis();
  
              return;
  
          }
  
          if (RECORD_LENGTH > 0) {
  
              int i = imagesIndex++ % images.length;
  
              yuvImage = images[i];
  
              timestamps[i] = 1000 * (System.currentTimeMillis() - startTime);
  
          }
  
          /* get video data */
  
          imageWidth = 640;
  
          imageHeight = 480    
  
          int finalImageHeight = 360;
  
          if (yuvImage != null && recording) {
  
              ByteBuffer bb = (ByteBuffer)yuvImage.image[0].position(0); // resets the buffer
  
              final int startY = imageWidth*(imageHeight-finalImageHeight)/2;
  
              final int lenY = imageWidth*finalImageHeight;
  
              bb.put(data, startY, lenY);
  
              final int startVU = imageWidth*imageHeight + imageWidth*(imageHeight-finalImageHeight)/4;
  
              final int lenVU = imageWidth* finalImageHeight/2;
  
              bb.put(data, startVU, lenVU);
  
              try {
  
                  long t = 1000 * (System.currentTimeMillis() - startTime);
  
                  if (t > recorder.getTimestamp()) {
  
                      recorder.setTimestamp(t);
  
                  }
  
                  recorder.record(yuvImage);
  
              } catch (FFmpegFrameRecorder.Exception e) {
  
                  Log.e(LOG_TAG, "problem with recorder():", e);
  
              }
  
          }
  
  
  
  
  
      }
  
  }

  Please also note that this solution was provided for an older version of javacv. The resulting video had a yellowish overlay covering 2/3rd part. Also there was empty section on left side as the video was not cropped correctly.

  So my question is what is the most appropriate solution for cropping videos using latest version of javacv ?

  Code after making change as suggested by Alex Cohn

      @Override
  
      public void onPreviewFrame(byte[] data, Camera camera) {
  
          if (audioRecord == null || audioRecord.getRecordingState() != AudioRecord.RECORDSTATE_RECORDING) {
  
              startTime = System.currentTimeMillis();
  
              return;
  
          }
  
          if (RECORD_LENGTH > 0) {
  
              int i = imagesIndex++ % images.length;
  
              yuvImage = images[i];
  
              timestamps[i] = 1000 * (System.currentTimeMillis() - startTime);
  
          }
  
          /* get video data */
  
          imageWidth = 640;
  
          imageHeight = 480;       
  
          destWidth = 480;
  
  
  
          if (yuvImage != null && recording) {
  
              ByteBuffer bb = (ByteBuffer)yuvImage.image[0].position(0); // resets the buffer
  
              int start = 2*((imageWidth-destWidth)/4); // this must be even
  
              for (int row=0; row2; row++) {
  
                  bb.put(data, start, destWidth);
  
                  start += imageWidth;
  
              }
  
              try {
  
                  long t = 1000 * (System.currentTimeMillis() - startTime);
  
                  if (t > recorder.getTimestamp()) {
  
                      recorder.setTimestamp(t);
  
                  }
  
                  recorder.record(yuvImage);
  
              } catch (FFmpegFrameRecorder.Exception e) {
  
                  Log.e(LOG_TAG, "problem with recorder():", e);
  
              }
  
          }
  
  
  
  
  
      }

  Screen shot from video generated with this code (destWidth 480) is

  

  Next I tried capturing a video with destWidth speciified as 639. The result is

  

  When destWidth is 639 video is repeating contents twice. When it is 480, contents are repeated 5 times and the green overlay and distortion is more.

  Also When the destWidth = imageWidth, video is captured properly. ie, for 640*480 there is no repetition of video contents and no green overlay.

  Converting frame to IplImage

  When this question was asked first, I missed to mention that the record method in FFmpegFrameRecorder is now accepting object of type Frame whereas earlier it was IplImage object. So I tried to apply Alex Cohn’s solution by converting Frame to IplImage.

  //---------------------------------------
  
  // initialize ffmpeg_recorder
  
  //---------------------------------------
  
  private void initRecorder() {
  
  
  
      Log.w(LOG_TAG,"init recorder");
  
  
  
      imageWidth = 640;
  
      imageHeight = 480; 
  
  
  
      if (RECORD_LENGTH > 0) {
  
          imagesIndex = 0;
  
          images = new Frame[RECORD_LENGTH * frameRate];
  
          timestamps = new long[images.length];
  
          for (int i = 0; i < images.length; i++) {
  
              images[i] = new Frame(imageWidth, imageHeight, Frame.DEPTH_UBYTE, 2);
  
              timestamps[i] = -1;
  
          }
  
      } else if (yuvImage == null) {
  
          yuvImage = new Frame(imageWidth, imageHeight, Frame.DEPTH_UBYTE, 2);
  
          Log.i(LOG_TAG, "create yuvImage");
  
          OpenCVFrameConverter.ToIplImage converter = new OpenCVFrameConverter.ToIplImage();
  
          yuvIplimage = converter.convert(yuvImage);
  
  
  
      }
  
  
  
      Log.i(LOG_TAG, "ffmpeg_url: " + ffmpeg_link);
  
      recorder = new FFmpegFrameRecorder(ffmpeg_link, imageWidth, imageHeight, 1);
  
      recorder.setFormat("flv");
  
      recorder.setSampleRate(sampleAudioRateInHz);
  
      // Set in the surface changed method
  
      recorder.setFrameRate(frameRate);
  
  
  
      Log.i(LOG_TAG, "recorder initialize success");
  
  
  
      audioRecordRunnable = new AudioRecordRunnable();
  
      audioThread = new Thread(audioRecordRunnable);
  
      runAudioThread = true;
  
  }
  
  
  
  
  
  
  
  @Override
  
      public void onPreviewFrame(byte[] data, Camera camera) {
  
          if (audioRecord == null || audioRecord.getRecordingState() != AudioRecord.RECORDSTATE_RECORDING) {
  
              startTime = System.currentTimeMillis();
  
              return;
  
          }
  
          if (RECORD_LENGTH > 0) {
  
              int i = imagesIndex++ % images.length;
  
              yuvImage = images[i];
  
              timestamps[i] = 1000 * (System.currentTimeMillis() - startTime);
  
          }
  
          /* get video data */
  
          int destWidth = 640;
  
  
  
          if (yuvIplimage != null && recording) {
  
              ByteBuffer bb = yuvIplimage.getByteBuffer(); // resets the buffer
  
              int start = 2*((imageWidth-destWidth)/4); // this must be even
  
              for (int row=0; row2; row++) {
  
                  bb.put(data, start, destWidth);
  
                  start += imageWidth;
  
              }
  
              try {
  
                  long t = 1000 * (System.currentTimeMillis() - startTime);
  
                  if (t > recorder.getTimestamp()) {
  
                      recorder.setTimestamp(t);
  
                  }
  
                  recorder.record(yuvImage);
  
              } catch (FFmpegFrameRecorder.Exception e) {
  
                  Log.e(LOG_TAG, "problem with recorder():", e);
  
              }
  
          }
  
  
  
  
  
      }

  But the videos generated with this method contained only green frames.

• How to encode Planar 4:2:0 (fourcc P010)

  20 juillet 2021, par DennisFleurbaaij

  I'm trying to recode fourcc V210 (which is a packed YUV4:2:2 format) into a P010 (planar YUV4:2:0). I think I've implemented it according to spec, but the renderer is giving a wrong image so something is off. Decoding the V210 has a decent example in ffmpeg (defines are modified from their solution) but I can't find a P010 encoder to look at what I'm doing wrong.

  


  (Yes, I've tried ffmpeg and that works but it's too slow for this, it takes 30ms per frame on an Intel Gen11 i7)

  


  Clarification (after @Frank's question) : The frames being processed are 4k (3840px wide) and hence there is no code for doing the 128b alignment.

  


  This is running on intel so little endian conversions applied.

  


  Try1 - all green image :

  


  The following code

  


  #define V210_READ_PACK_BLOCK(a, b, c) \
    do {                              \
        val  = *src++;                \
        a = val & 0x3FF;              \
        b = (val >> 10) & 0x3FF;      \
        c = (val >> 20) & 0x3FF;      \
    } while (0)

#define PIXELS_PER_PACK 6
#define BYTES_PER_PACK (4*4)

void MyClass::FormatVideoFrame(
    BYTE* inFrame,
    BYTE* outBuffer)
{
    const uint32_t pixels = m_height * m_width;

    const uint32_t* src = (const uint32_t *)inFrame);

    uint16_t* dstY = (uint16_t *)outBuffer;

    uint16_t* dstUVStart = (uint16_t*)(outBuffer + ((ptrdiff_t)pixels * sizeof(uint16_t)));
    uint16_t* dstUV = dstUVStart;

    const uint32_t packsPerLine = m_width / PIXELS_PER_PACK;

    for (uint32_t line = 0; line < m_height; line++)
    {
        for (uint32_t pack = 0; pack < packsPerLine; pack++)
        {
            uint32_t val;
            uint16_t u, y1, y2, v;

            if (pack % 2 == 0)
            {
                V210_READ_PACK_BLOCK(u, y1, v);
                *dstUV++ = u;
                *dstY++ = y1;
                *dstUV++ = v;

                V210_READ_PACK_BLOCK(y1, u, y2);
                *dstY++ = y1;
                *dstUV++ = u;
                *dstY++ = y2;

                V210_READ_PACK_BLOCK(v, y1, u);
                *dstUV++ = v;
                *dstY++ = y1;
                *dstUV++ = u;

                V210_READ_PACK_BLOCK(y1, v, y2);
                *dstY++ = y1;
                *dstUV++ = v;
                *dstY++ = y2;
            }
            else
            {
                V210_READ_PACK_BLOCK(u, y1, v);
                *dstY++ = y1;

                V210_READ_PACK_BLOCK(y1, u, y2);
                *dstY++ = y1;
                *dstY++ = y2;

                V210_READ_PACK_BLOCK(v, y1, u);
                *dstY++ = y1;

                V210_READ_PACK_BLOCK(y1, v, y2);
                *dstY++ = y1;
                *dstY++ = y2;
            }
        }
    }

#ifdef _DEBUG

    // Fully written Y space
    assert(dstY == dstUVStart);

    // Fully written UV space
    const BYTE* expectedVurrentUVPtr = outBuffer + (ptrdiff_t)GetOutFrameSize();
    assert(expectedVurrentUVPtr == (BYTE *)dstUV);

#endif
}

// This is called to determine outBuffer size
LONG MyClass::GetOutFrameSize() const
{
    const LONG pixels = m_height * m_width;

    return
        (pixels * sizeof(uint16_t)) +  // Every pixel 1 y
        (pixels / 2 / 2 * (2 * sizeof(uint16_t)));  // Every 2 pixels and every odd row 2 16-bit numbers
}


  


  Leads to all green image. This turned out to be a missing bit shift to place the 10 bits in the upper bits of the 16-bit value as per the P010 spec.

  


  Try 2 - Y works, UV doubled ?

  


  Updated the code to properly (or so I think) shifts the YUV values to the correct position in their 16-bit space.

  


  #define V210_READ_PACK_BLOCK(a, b, c) \
    do {                              \
        val  = *src++;                \
        a = val & 0x3FF;              \
        b = (val >> 10) & 0x3FF;      \
        c = (val >> 20) & 0x3FF;      \
    } while (0)


#define P010_WRITE_VALUE(d, v) (*d++ = (v << 6))

#define PIXELS_PER_PACK 6
#define BYTES_PER_PACK (4 * sizeof(uint32_t))

// Snipped constructor here which guarantees that we're processing
// something which does not violate alignment.

void MyClass::FormatVideoFrame(
    const BYTE* inBuffer,
    BYTE* outBuffer)
{   
    const uint32_t pixels = m_height * m_width;
    const uint32_t aligned_width = ((m_width + 47) / 48) * 48;
    const uint32_t stride = aligned_width * 8 / 3;

    uint16_t* dstY = (uint16_t *)outBuffer;

    uint16_t* dstUVStart = (uint16_t*)(outBuffer + ((ptrdiff_t)pixels * sizeof(uint16_t)));
    uint16_t* dstUV = dstUVStart;

    const uint32_t packsPerLine = m_width / PIXELS_PER_PACK;

    for (uint32_t line = 0; line < m_height; line++)
    {
        // Lines start at 128 byte alignment
        const uint32_t* src = (const uint32_t*)(inBuffer + (ptrdiff_t)(line * stride));

        for (uint32_t pack = 0; pack < packsPerLine; pack++)
        {
            uint32_t val;
            uint16_t u, y1, y2, v;

            if (pack % 2 == 0)
            {
                V210_READ_PACK_BLOCK(u, y1, v);
                P010_WRITE_VALUE(dstUV, u);
                P010_WRITE_VALUE(dstY, y1);
                P010_WRITE_VALUE(dstUV, v);

                V210_READ_PACK_BLOCK(y1, u, y2);
                P010_WRITE_VALUE(dstY, y1);
                P010_WRITE_VALUE(dstUV, u);
                P010_WRITE_VALUE(dstY, y2);

                V210_READ_PACK_BLOCK(v, y1, u);
                P010_WRITE_VALUE(dstUV, v);
                P010_WRITE_VALUE(dstY, y1);
                P010_WRITE_VALUE(dstUV, u);

                V210_READ_PACK_BLOCK(y1, v, y2);
                P010_WRITE_VALUE(dstY, y1);
                P010_WRITE_VALUE(dstUV, v);
                P010_WRITE_VALUE(dstY, y2);
            }
            else
            {
                V210_READ_PACK_BLOCK(u, y1, v);
                P010_WRITE_VALUE(dstY, y1);

                V210_READ_PACK_BLOCK(y1, u, y2);
                P010_WRITE_VALUE(dstY, y1);
                P010_WRITE_VALUE(dstY, y2);

                V210_READ_PACK_BLOCK(v, y1, u);
                P010_WRITE_VALUE(dstY, y1);

                V210_READ_PACK_BLOCK(y1, v, y2);
                P010_WRITE_VALUE(dstY, y1);
                P010_WRITE_VALUE(dstY, y2);
            }
        }
    }

#ifdef _DEBUG

    // Fully written Y space
    assert(dstY == dstUVStart);

    // Fully written UV space
    const BYTE* expectedVurrentUVPtr = outBuffer + (ptrdiff_t)GetOutFrameSize();
    assert(expectedVurrentUVPtr == (BYTE *)dstUV);

#endif
}


  


  This leads to the Y being correct and the amount of lines for U and V as well, but somehow U and V are not overlaid properly. There are two versions of it seemingly mirrored through the center vertical. Something similar but less visible for zeroing out V. So both of these are getting rendered at half the width ? Any tips appreciated :)

  


  Fix :
Found the bug, I'm flipping VU not per pack but per block

  


  if (pack % 2 == 0)


  


  Should be

  


  if (line % 2 == 0)