Recherche avancée

Sur d’autres sites (13028)

• Preprocessing of Columbia active speaker dataset

  21 octobre 2019, par aspit794

  I would like to use the Columbia dataset (which is 1 video https://youtu.be/6GzxbrO0DHM) and its ground truth annotation (http://www.jaychakravarty.com/wp-content/uploads/2016/02/columbiaDatasetGroundTruth.zip).

  annotation contain the frame_id (int), x and y position of the bounding box (int, int), the width=height of the bounding box (int), and the ground truth for active speaker (0 or 1)

  More information : https://www.jaychakravarty.com/active-speaker-detection/

  I managed to download the video, extracted the frames with

  ffmpeg -loglevel 0 -i columbia.mkv columbia/frame_%06d.jpg -nostdin -vf -an -hide_banner

  and draw the ground truth bounding boxes to the video frames (where available).

  However, the bounding boxes are not in the correct places, and I am a bit helpless.

  I tried to convert the video to other frame rate (e.g. 25 fps), but the annotation is not fit.

  I read every paper available, which is related to this dataset, but neither of them mention any information about preprocessing, and the codebase is not publicly available.

  How should I preprocess the video to interpret the annotation ?

• avcodec/cavsdec : Check remaining bitstream in the main loop in decode_pic()

  24 septembre 2019, par Michael Niedermayer

  avcodec/cavsdec : Check remaining bitstream in the main loop in decode_pic()
  Fixes : Timeout (149sec ->1sec)
  
  Fixes : 17311/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_CAVS_fuzzer-5679368642232320
  Found-by : continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
  
  Signed-off-by : Michael Niedermayer <michael@niedermayer.cc>
  

  • [DH] libavcodec/cavsdec.c
  • [DH] tests/ref/fate/cavs

• FFMPEG amix filter causes main audio stream to cut out

  21 janvier 2019, par user3600107

  So I have a video called 1.mkv and would like to mix in a variety of different audio clips at certain points. To do this I’m using the -filter_complex option. However, I’m running into some problems because when ffmpeg tries to mix in the first audio stream the audio works for a short while when the clip is playing and then all audio cuts out. I’m running ffmpeg version 2.8.15-0(which is up to date with my distro). Another "weird" thing about the video output is that in xplayer the video will freeze after the audio cuts out, and will work if you skip far enough ahead in the video(not sure if this is helpful but it might give some extra clues).

  Full command :

  ffmpeg -i "1.mkv" -i "5.wav" -i "2.wav" -i "3.wav" -i "6.wav" -i "7.wav" -i "4.wav" -i "9.wav" -i "8.wav" -i "10.wav" -filter_complex "[0:0]setdar=4/3[v0];
  
  
  
  [2:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=20000|20000,volume=0.5[ad2];
  
  
  
  [4:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=900000|900000,volume=0.5[ad4];
  
  
  
  [3:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=300000|300000,volume=0.5[ad3];
  
  
  
  [1:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=720000|720000,volume=0.5[ad1];
  
  
  
  [7:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=1140000|1140000,volume=0.5[ad7];
  
  
  
  [9:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=1260000|1260000,volume=0.5[ad9];
  
  
  
  [8:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=1020000|1020000,volume=0.5[ad8];
  
  
  
  [5:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=960000|960000,volume=0.5[ad5];
  
  
  
  [6:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=480000|480000,volume=0.5[ad6];
  
  [0:1]volume=1[ad0];
  
  [ad0][ad2][ad4][ad3][ad1][ad7][ad9][ad8][ad5][ad6]amix=inputs=10:duration=first:dropout_transition=0,dynaudnorm[a0]" -map "[v0]" -map "[a0]" -c:v libx264 -ar 44100  -c:a libmp3lame -preset ultrafast -crf 17 -b:v 1M out2.flv

  partial command

  ffmpeg -i "1.mkv" -i "2.wav" -filter_complex "[0:0]setdar=4/3[v0];
  
  [1:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay=20000|20000,volume=0.5[ad2];
  
  [0:1]volume=1[ad0];
  
  [ad0][ad2]amix=inputs=2:duration=first:dropout_transition=0,dynaudnorm[a0]" -map "[v0]" -map "[a0]" -c:v libx264 -ar 44100  -c:a libmp3lame -preset ultrafast -crf 17 -b:v 1M out2.flv