Recherche avancée

Sur d’autres sites (9786)

• aarch64 : h264qpel : Do vertical filtering without transposing

  3 septembre 2021, par Martin Storsjö

  aarch64 : h264qpel : Do vertical filtering without transposing
  This gives rather big speedups on these functions :
  Before :
  
  put_h264_qpel_8_mc01_8_neon :     241.0   131.5   138.7
  
  put_h264_qpel_8_mc02_8_neon :     214.7   121.2   127.5
  
  put_h264_qpel_8_mc03_8_neon :     242.5   131.2   135.7
  
  put_h264_qpel_8_mc11_8_neon :     421.2   218.7   251.0
  
  put_h264_qpel_8_mc12_8_neon :     878.0   509.5   537.5
  
  put_h264_qpel_8_mc13_8_neon :     423.7   217.0   252.0
  
  put_h264_qpel_8_mc21_8_neon :     858.2   479.5   514.0
  
  put_h264_qpel_8_mc22_8_neon :     649.7   385.2   403.0
  
  put_h264_qpel_8_mc23_8_neon :     860.2   476.5   517.7
  
  put_h264_qpel_8_mc31_8_neon :     437.2   219.5   252.5
  
  put_h264_qpel_8_mc32_8_neon :     892.5   510.5   546.0
  
  put_h264_qpel_8_mc33_8_neon :     438.2   218.5   257.0
  
  put_h264_qpel_16_mc01_8_neon :    944.2   509.7   546.7
  
  put_h264_qpel_16_mc02_8_neon :    878.7   469.5   509.7
  
  put_h264_qpel_16_mc03_8_neon :    945.7   510.7   557.0
  
  put_h264_qpel_16_mc11_8_neon :   1663.2   858.5   979.5
  
  put_h264_qpel_16_mc12_8_neon :   3510.2  2027.7  2112.7
  
  put_h264_qpel_16_mc13_8_neon :   1664.7   857.5   980.5
  
  put_h264_qpel_16_mc21_8_neon :   3366.2  1928.5  2030.5
  
  put_h264_qpel_16_mc22_8_neon :   2584.7  1514.7  1590.2
  
  put_h264_qpel_16_mc23_8_neon :   3367.7  1927.7  2035.0
  
  put_h264_qpel_16_mc31_8_neon :   1716.7   849.7   997.0
  
  put_h264_qpel_16_mc32_8_neon :   3564.0  2044.2  3835.2
  
  put_h264_qpel_16_mc33_8_neon :   1717.7   863.0   989.5
  After :
  
  put_h264_qpel_8_mc01_8_neon :     136.0    73.7    76.0
  
  put_h264_qpel_8_mc02_8_neon :     108.7    65.0    64.0
  
  put_h264_qpel_8_mc03_8_neon :     137.5    72.7    73.0
  
  put_h264_qpel_8_mc11_8_neon :     316.2   159.0   188.5
  
  put_h264_qpel_8_mc12_8_neon :     653.0   375.5   384.7
  
  put_h264_qpel_8_mc13_8_neon :     318.7   165.5   189.5
  
  put_h264_qpel_8_mc21_8_neon :     739.2   385.7   432.5
  
  put_h264_qpel_8_mc22_8_neon :     530.7   295.5   309.5
  
  put_h264_qpel_8_mc23_8_neon :     741.2   393.7   421.0
  
  put_h264_qpel_8_mc31_8_neon :     332.2   162.5   190.0
  
  put_h264_qpel_8_mc32_8_neon :     667.5   378.2   390.5
  
  put_h264_qpel_8_mc33_8_neon :     332.7   166.5   195.5
  
  put_h264_qpel_16_mc01_8_neon :    524.2   285.2   294.0
  
  put_h264_qpel_16_mc02_8_neon :    454.7   252.2   250.2
  
  put_h264_qpel_16_mc03_8_neon :    525.7   286.0   283.0
  
  put_h264_qpel_16_mc11_8_neon :   1243.2   630.7   726.7
  
  put_h264_qpel_16_mc12_8_neon :   2610.2  1479.7  1481.2
  
  put_h264_qpel_16_mc13_8_neon :   1250.5   631.7   727.7
  
  put_h264_qpel_16_mc21_8_neon :   2890.2  1571.2  1679.7
  
  put_h264_qpel_16_mc22_8_neon :   2108.7  1177.5  1223.5
  
  put_h264_qpel_16_mc23_8_neon :   2891.7  1578.7  1667.7
  
  put_h264_qpel_16_mc31_8_neon :   1296.7   630.5   752.5
  
  put_h264_qpel_16_mc32_8_neon :   2664.0  1483.2  1503.5
  
  put_h264_qpel_16_mc33_8_neon :   1297.7   632.5   747.2
  I.e. overall a 20%-60% reduction in runtime of these
  
  functions.
  Signed-off-by : Martin Storsjö <martin@martin.st>
  

  • [DH] libavcodec/aarch64/h264qpel_neon.S

• x86/aacencdsp : add AVX version of quantize_bands

  1er juin 2024, par James Almer

  x86/aacencdsp : add AVX version of quantize_bands
  quant_bands_signed_c : 1928.0
  
  quant_bands_signed_sse2 : 406.0
  
  quant_bands_signed_avx : 207.0
  
  quant_bands_unsigned_c : 1702.0
  
  quant_bands_unsigned_sse2 : 404.0
  
  quant_bands_unsigned_avx : 209.0
  Signed-off-by : James Almer <jamrial@gmail.com>
  

  • [DH] libavcodec/aacenc.h
  • [DH] libavcodec/x86/aacencdsp.asm
  • [DH] libavcodec/x86/aacencdsp_init.c
  • [DH] tests/checkasm/aacencdsp.c

• Live stream is gets delayed while processing frame in opencv + python

  18 mars 2021, par Himanshu sharma

  I capture and process an IP camera RTSP stream in a OpenCV 4.4.0.46 on Ubuntu.
Unfortunately the processing takes quite a lot of time, roughly 0.2s per frame, and the stream quickly gets delayed.
Video file have to save for 5 min but by this delaying video file is saved for 3-4 min only.

  


  Can we process faster to overcome delays ?

  


  I have two IP camera which have two diffrent fps_rate(Camera 1 have 18000 and camera 2 have 20 fps)

  


  I am implementing this code in difference Ubuntu PCs

  


  

  • Python 3.8.5 (default, Jul 28 2020, 12:59:40) [GCC 9.3.0] on linux

  


  • Django==3.1.2

  


  • Ubuntu = 18.04 and 20.04

  


  • opencv-contrib-python==4.4.0.46

  


  • opencv-python==4.4.0.46

  


  


  input_stream = 'rtsp://'+username+':'+password+'@'+ip+'/user='+username+'_password='+password+'_channel=0channel_number_stream=0.sdp'
input_stream---> rtsp://admin:Admin123@192.168.1.208/user=admin_password=Admin123_channel=0channel_number_stream=0.sdp

input_stream---> rtsp://Admin:@192.168.1.209/user=Admin_password=_channel=0channel_number_stream=0.sdp

vs = cv2.VideoCapture(input_stream)
fps_rate = int(vs.get(cv2.CAP_PROP_FPS))
I have two IP camera which have two diffrent fps_rate(Camera 1 have 18000 and camera 2 have 20 fps)

video_file_name = 0
start_time = time.time()
while(True):
    ret, frame = vs.read()
    time.sleep(0.2)     # <= Simulate processing time (mask detection, face detection and many detection is hapning)


    ###  Start of  writing a video to disk          
    minute = 5  ## saving a file for 5 minute only then saving another file for 5 min
    second  = 60
    minite_to_save_video = int(minute) * int(second)


    # if we are supposed to be writing a video to disk, initialize
    if time.time() - start_time >= minite_to_save_video or  video_file_name == 0 :
        ## where H = heigth, W = width, C = channel 
        H, W, C = frame.shape
        
        print('time.time()-->',time.time(),'video_file_name-->', video_file_name,  ' #####')
        start_time = time.time()

        video_file_name = str(time.mktime(datetime.datetime.now().timetuple())).replace('.0', '')
        output_save_directory = output_stream+str(int(video_file_name))+'.mp4'


        fourcc = cv2.VideoWriter_fourcc(*'avc1')
        
        writer = cv2.VideoWriter(output_save_directory, fourcc,20.0,(W, H), True)

    # check to see if we should write the frame to disk
    if writer is not None:
        
        try:
            writer.write(frame)

        except Exception as e:
            print('Error in writing video output---> ', e)