Recherche avancée

Sur d’autres sites (11728)

• How to encode a HEVC video from YUV420 Mat data (from a hardware triggered camera) via FFmpeg in Python

  4 juillet 2023, par QuantumRiver

  I am having problems encoding a HEVC video from a series of YUV420 Mat data via FFmpeg.

  


  

  1. I am using python in Ubuntu-20.04 ;

  


  2. I am retrieving frame data from a hardware triggered camera (BASLER), using pypylon ;

  


  3. I want to write a video from that camera in HEVC codec, using my GPU-NVENC ;

  


  4. I guess I have to use FFmpeg to achieve these ;

  


  


  What I have tried :

  


  

  1. I find that FFmpeg supports encoding from a camera, but it seems to only support webcams, not the camera I use (hardware triggered BASLER cameras with pypylon APIs) ;

  


  2. I find that FFmpeg supports transfering a video from one codec to another, which is not my case ;

  


  3. I find that FFmpeg supports encoding a video from a series of jpeg images. But in my case, it will be inefficient if I first save each frame into a picture and then encode them into a video ;

  


  4. The frame data retrieved from camera can be converted to YUV420 (directly from pypylon), which is suitable for HEVC encoding ;

  


  5. I learnt that the basic unit in FFmpeg to encode a video is AVFrame. I guess I have to first turn my YUV420 data into AVFrame, then encode AVFrames into HEVC ;

  


  6. But I do not know how to achieve that in python.

  


  


  My simplified and expected codes :

  


  camera = pylon.InstantCamera(tlf.CreateFirstDevice())
converter = pylon.ImageFormatConverter()
converter.OutputPixelFormat = pylon.PixelType_YUV420
video_handle = xxxxxx  # HEVC
while True:
    grabResult = camera.RetrieveResult(timeout, pylon.TimeoutHandling_ThrowException)
    image = converter.Convert(grabResult).GetArray()
    video_handle.write(frame)  # encode into a hevc video via ffmpeg in NVENC


  


• FFMPEG : Reversing only a segment of a video file

  25 juillet 2019, par N. Johnson

  I am writing a script that takes a piece of video and then reverses the same piece of video and then concats the two together so the final video plays forwards and then loops backwards. I should note that eventually I want to be able to pull an unequal length for the reverse part.

  I can get the entire file to do this, but getting just a segment is not working as expected.

  See code below

  I’ve tried :

  %1 = timecode to seek to (the video file is only 20 seconds and never any longer)

  %2 = length of segment to pull out

  %3 = usually the same as %2 but may be different if we want to only reverse 2 seconds instead of the full 5 for example.

  ffmpeg -ss 00:00:%1 -an -i test.mp4 -t 00:00:%2 out.mp4
  
  :: the above works as expected
  
  
  
  :: this doesn't, no matter what I put into -ss. I've also tried moving -ss out front of the -i as suggested in the documentation? It gives me the right length of segment but never starts in the right place. 
  
  
  
  ffmpeg  -an -i test.mp4 -ss 00:00:xx -t 00:00:%3 -vf reverse reversed2.mp4
  
  
  
  :the below works fine
  
  ffmpeg -f concat -safe 0 -i mylist.txt -c copy output.mp4

  When I run this with say %1 = 05 and %2 = 05, I get a segment from 5 seconds in that lasts 5 seconds. Then I get a seemingly random starting point and 5 seconds of reversed video. I’ve tried a number of inputs in "XX" from 10 (which I think is right) to 0 to 19 and all of them produce output. But it’s all wrong.

• Correct way to dump yuv values after avcodec_decode_video2

  27 juin 2017, par Peter

  I am decoding H.265 raw data :

   avcodec_decode_video2(decoderCtx, pictYUV, &gotPicture, &packet)

  After the call, the value for pictYUV->format is AV_PIX_FMT_YUV420P as expected.

  I dump raw yuv values to a file as follows :

  fwrite(pictYUV->data[0], 1, w*h, f);
  
  fwrite(pictYUV->data[2], 1, w*h/4, f);
  
  fwrite(pictYUV->data[1], 1, w*h/4, f);

  For an input size of 640x480, when the file is viewed (using ImageMagick display utility), the image is what was expected.

  However, for an input size of 864x480, the image appears to be corrupted.

  What is interesting is if I run pictYUV through sws_scale, and dump the resulting yuv output, the image appears to be fine.

  sws_scale(swsCtx, pictYUV->data, pictYUV->linesize, 0, pictYUV->height,
  
    pictNewYUV->data, pictNewYUV->linesize);

  All I need is yuv data. I am hoping I can avoid the extra call to sws_scale. Wondering what is it that I am missing. Regards.