Recherche avancée

Sur d’autres sites (5513)

• How to extract frame types along with motion vectors using extract_mvs.c from ffmpeg

  26 février 2018, par helmo

  I have been researching ways to get frame types (I, P, B) along with the motion vector data returned from extract_mvs.c in the examples folder in ffmpeg.

  The extract_mvs.c file after it is compiled, returns information like this :

  framenum,source,blockw,blockh,srcx,srcy,dstx,dsty,flags
  
  2,-1,16,16,   8,   8,   8,   8,0x0
  
  2, 1,16,16,   8,   8,   8,   8,0x0
  
  2, 1,16,16,  24,   8,  24,   8,0x0
  
  2, 1,16,16,  40,   8,  40,   8,0x0
  
  2, 1,16,16,  56,   8,  56,   8,0x0
  
  2, 1,16,16,  72,   8,  72,   8,0x0
  
  2, 1,16,16,  88,   8,  88,   8,0x0
  
  ...
  
  297, 1,16,16, 248, 280, 248, 280,0x0
  
  297, 1,16,16, 264, 280, 264, 280,0x0
  
  297,-1,16,16, 278, 279, 280, 280,0x0
  
  297, 1,16,16, 280, 280, 280, 280,0x0
  
  297, 1,16,16, 296, 280, 296, 280,0x0
  
  297, 1,16,16, 312, 280, 312, 280,0x0
  
  297, 1,16,16, 328, 280, 328, 280,0x0
  
  297, 1,16,16, 344, 280, 344, 280,0x0

  Along with this information, I would like to output frame type so that I know framenum = 2 is, for example, a ’B’ frame.

  I tried different things, one of which was using a separate command :

  ffprobe input.mp4 -show_frames | grep -E 'pict_type|coded_picture_number'

  But the problem with this command is that it returns data like :

  pict_type=I
  
  coded_picture_number=0
  
  pict_type=B
  
  coded_picture_number=2
  
  pict_type=P
  
  coded_picture_number=1
  
  pict_type=B
  
  coded_picture_number=4
  
  pict_type=P
  
  coded_picture_number=3
  
  ....
  
  pict_type=P
  
  coded_picture_number=293
  
  pict_type=B
  
  coded_picture_number=297
  
  pict_type=B
  
  coded_picture_number=296

  And there is no much I can relate here between coded_picture_number and framenum. The former starts counting from 0 and the later from 2. I assume framenum starting from 2, means the count from this variable is actually from 1, and it ignored 1 in the extraction process as it is maybe an I frame thus no motion vectors.

  So, how can we use only extract_mvs.c to get not only that information it provides but also the frame types in the returned table. Any hints either syntax/command-wise or in editing the c file would be appreciated. Thanks in advance.

• RGB to YUV422 conversion with ffmpeg, incorrect colors

  22 janvier 2016, par user3578571

  I’m trying to convert an 8bit RGB uncompressed to an mpeg2 mxf file (xdcam 422 HD 1080 50i) which is YUV422. With info from the FFMpeg docs and various websites i made the following command :

  ./ffmpeg -y -i test_lines.mov  -pix_fmt yuv422p -vcodec mpeg2video -non_linear_quant 1 -flags +ildct+ilme -top 1 -dc 10 -intra_vlc 1 -qmax 2 -vtag xd5c -rc_max_vbv_use 1 -rc_min_vbv_use 1 -g 12 -b:v 50000k -minrate 50000k -maxrate 50000k -bufsize 8000k -acodec pcm_s24le -ar 48000 -bf 2 -ac 2 lines_HD.mxf

  This gave me a result with the colors much brighter than the original.

  So i tried adding the options -color_range 1 -colorspace 1 -color_primaries 1 -color_trc 1 but this didn’t seem to do anything.

  After adding colormatrix=bt601:bt709 i got a way better image, but slightly darker than the original and it also feels weird specifying this option cause the source is also in the REC709 colorspace, so why specify it differently ?

  Next i regenerated my source image to an YUV codec (prores) and rerun FFMpeg on it with the colors coming out just fine. Therefore i think it has to be an RGB -> YUV problem.
  Does somebody have an idea how to this properly ? I can provide screenshots of the different results on a videoscope as soon as i’m back at the office, if anybody is interested.

  Last, i know there are various topics touching this subject but either they go way over my head FFmpeg wise or bring me to the stage where i already am.

• FFmpeg/Libav LZW compression

  1er avril 2016, par Sir DrinksCoffeeALot

  I’m trying to implement libav’s LZW compression and decompression methods into my own project since the code is already there and i don’t have to invent anything new (i’m using LZW just to test how much can i compress video before sending it through LAN).

  I’m not quite sure if i’m doing something wrong since their code is quite complex (at some point they are shifting bits and such), and some parts i can’t really understand. But anyways, here is what i have tried so far.

  I’ve allocated/initialized few buffers (input/output), LZWEncodeState and LZWState structures. After that i’ve called function that fills up LZWEncodeState structure with some default values (ff_lzw_encode_init), and after that i called ff_lzw_encode and passed LZWEncodeState reference together with input buffer and it’s size. That’s what iv’e done encoding-part wise.

  For the decode part, we have to call ff_lzw_encode_open which basicly allocates LZWState structure memory, ff_lzw_decode_init (fills LZWState with default/passed valued) and after that, we use ff_lzw_decode to decode buffer back into original one (since LZW should be lossless).

  The thing is that after encoding/decoding, in output buffer only first 2 values are same as those in input buffer. I have no idea why is that happening. I’ve tried different values, but nothing has changed. So if anyone else experienced these kind of problems, and figured how to solve them, please let me know or try to explain why is that happening.