Recherche avancée

Sur d’autres sites (7418)

• FFMPEG/tensorflow warping .mp4's with no observable pattern of correlation

  25 mai 2018, par ThePeskyWabbit

  The function that performs the processing is here :

  def ffwd_video(path_in, path_out, checkpoint_dir, device_t='/gpu:0', batch_size=4):
  
  video_clip = VideoFileClip(path_in, audio=False)
  
  video_writer = ffmpeg_writer.FFMPEG_VideoWriter(path_out, video_clip.size, video_clip.fps, codec="libx264",
  
                                                  preset="slow", bitrate="3000k",
  
                                                  audiofile=path_in, def ffwd_video(path_in, path_out, checkpoint_dir, device_t='/gpu:0', batch_size=2):threads=None,
  
                                                  ffmpeg_params=None)
  
  
  
  g = tf.Graph()
  
  soft_config = tf.ConfigProto(allow_soft_placement=True)
  
  soft_config.gpu_options.allow_growth = True
  
  with g.as_default(), g.device(device_t), \
  
          tf.Session(config=soft_config) as sess:
  
  
  
      batch_shape = (batch_size, video_clip.size[1], video_clip.size[0], 3)
  
      print(batch_shape)
  
      img_placeholder = tf.placeholder(tf.float32, shape=batch_shape,
  
                                       name='img_placeholder')
  
  
  
      preds = transform.net(img_placeholder)
  
      print(preds)
  
      saver = tf.train.Saver()
  
      if os.path.isdir(checkpoint_dir):
  
          ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
  
          if ckpt and ckpt.model_checkpoint_path:
  
              saver.restore(sess, ckpt.model_checkpoint_path)
  
          else:
  
              raise Exception("No checkpoint found...")
  
      else:
  
          saver.restore(sess, checkpoint_dir)
  
  
  
      X = np.zeros(batch_shape, dtype=np.float32)
  
  
  
      def style_and_write(count):
  
          for i in range(count, batch_size):
  
              X[i] = X[count]  # Use last frame to fill X
  
          _preds = sess.run(preds, feed_dict={img_placeholder: X})
  
          for i in range(0, count):
  
              video_writer.write_frame(np.clip(_preds[i], 0, 255).astype(np.uint8))
  
  
  
      frame_count = 0  # The frame count that written to X
  
      for frame in video_clip.iter_frames():
  
          X[frame_count] = frame
  
          frame_count += 1
  
          if frame_count == batch_size:
  
              style_and_write(frame_count)
  
              frame_count = 0
  
  
  
      if frame_count != 0:
  
          style_and_write(frame_count)
  
  
  
      video_writer.close()

  I do not expect this to be a repeatable sample as it requires gigabytes of models/checkpoints/etc. to do so.

  Anyways, I am converting .gif’s to .mp4’s using this line of code :

  os.system('echo "y"| ffmpeg -i temp.gif -movflags faststart -pix_fmt yuv420p -vf "scale=trunc(iw/2)*2:trunc(ih/2)*2" temp.mp4')

  and when I run a subsequent .mp4 through the video rendering function, this will sometimes happen :

  original : https://i.imgur.com/WrBa6yw.gifv

  result : https://giphy.com/gifs/dream-bot-BzqB8XILJuVTJcgJnn

  It will happen on certain .mp4’s 100% of the time and on others, 0% of the time. It either happens to a clip or it does not happen to that clip. It is not a case of a single clip sometimes having this problem. If it is good, it is always good but if it’s bad, that clip will never work.

  After this output data from when the clips are processed :

  good gifs
  
  Stream #0:0: Video: h264 (libx264) ([33][0][0][0] / 0x0021), yuv420p, 728x728, q=-1--1, 30 fps, 15360 tbn, 30 tbc
  
  frame=  450 fps= 78 q=-1.0 Lsize=    7570kB time=00:00:14.93 bitrate=4152.6kbits/s
  
  
  
  Stream #0:0: Video: gif, bgra, 800x450, 25 fps, 25 tbr, 100 tbn, 100 tbc
  
  Stream #0:0: Video: h264 (libx264) ([33][0][0][0] / 0x0021), yuv420p, 728x408, q=-1--1, 23.98 fps, 24k tbn, 23.98 tbc
  
  frame=  360 fps=198 q=-1.0 Lsize=    2568kB time=00:00:14.93 bitrate=1408.8kbits/s 
  
  
  
  Stream #0:0: Video: gif, bgra, 728x408, 23.92 fps, 23.98 tbr, 100 tbn, 100 tbc
  
  Stream #0:0: Video: h264 (libx264) ([33][0][0][0] / 0x0021), yuv420p, 728x408, q=-1--1, 23.98 fps, 24k tbn, 23.98 tbc
  
  frame=  360 fps=198 q=-1.0 Lsize=    2568kB time=00:00:14.93 bitrate=1408.8kbits/s 
  
  
  
  bad gifs
  
  Stream #0:0: Video: h264 (libx264) ([33][0][0][0] / 0x0021), yuv420p, 800x450, q=-1--1, 25 fps, 12800 tbn, 25 tbc
  
  frame=  150 fps= 87 q=-1.0 Lsize=    1840kB time=00:00:05.92 bitrate=2545.8kbits/s
  
  
  
  Stream #0:0: Video: gif, bgra, 378x275, 25 fps, 25 tbr, 100 tbn, 100 tbc
  
  Stream #0:0: Video: h264 (libx264) ([33][0][0][0] / 0x0021), yuv420p, 378x274, q=-1--1, 25 fps, 12800 tbn, 25 tbc
  
  frame=  115 fps=0.0 q=-1.0 Lsize=     253kB time=00:00:04.52 bitrate= 459.4kbits/s

  I have noticed that the frame= value is exactly 15x larger than the .gif fps value while for the bad .gifs, it is not. I believe this is the cause but do not know how to change these values. does anybody know how I can process/convert the gif to .mp4 so that the frame= value is 15x the gif fps ?

• creating a video using ffmpeg command with image name pattern [duplicate]

  21 mars 2018, par Junaid Farooq

  This question already has an answer here :

  • ffmpeg img to video = Could find no file with path
    
    2 answers
    

  I am coming from this question
  https://askubuntu.com/questions/610903/how-can-i-create-a-video-file-from-a-set-of-jpg-images

  Everything is working fine and I can create a video, I am only struggling with the image name pattern. Each time, I had to rename all files first to a specific pattern.

  Can we create a video using the same command written below but with this kind of image name pattern ?

  2018-01-19-11-55-00.jpg  2018-01-19-17-00-00.jpg  2018-01-22-10-10-00.jpg  2018-01-22-13-25-00.jpg  2018-01-22-16-40-00.jpg
  
  2018-01-19-12-00-00.jpg  2018-01-19-17-05-00.jpg  2018-01-22-10-15-00.jpg  2018-01-22-13-30-00.jpg  2018-01-22-16-45-00.jpg

  I want to use this command

  ffmpeg -framerate 25 -i image-%04d.jpg -c:v libx264 -profile:v high -crf 20 -pix_fmt yuv420p output.mp4

  but this one using

  image-2000.jpg  image-2400.jpg  image-2800.jpg

  above image pattern.

  Is there any possibility to use the above pattern of files which are being named as a date. also, all these images are in a folder. Can FFMPEG address them ?

• avcodec/mips : preload data in hevc sao edge 135 degree filter msa functions

  9 octobre 2017, par Kaustubh Raste

  avcodec/mips : preload data in hevc sao edge 135 degree filter msa functions
  Signed-off-by : Kaustubh Raste <kaustubh.raste@imgtec.com>
  
  Reviewed-by : Manojkumar Bhosale <Manojkumar.Bhosale@imgtec.com>
  
  Signed-off-by : Michael Niedermayer <michael@niedermayer.cc>
  

  • [DH] libavcodec/mips/hevc_lpf_sao_msa.c