U (dM@sddlZddlZddlmZddlmZmZmZmZm Z ddl Z ddl m Z ze ddZ Wneefk rxdZ YnXedd ZGd d d ZGd d d ZeeefddddZe je je je je je jedddZe je jeeefe jdddZdddddddeddddef eeeeeeeeeefeeeeeeefeee je jefdddZeeeeeeefdddZeeddd Zd0e jeeeeeeeeefeeeeeeeefeeee je jfd!d"d#Ze jeeeeeefd$d%d&Ze jed$d'd(Z d1ee eefee eefeee je jeeeffd*d+d,Z!d2eeee eeeefeefd-d.d/Z"dS)3N)Fraction)ListTupleDictOptionalUnion) _load_library video_readerTFc@s0eZdZeedZddgZeeddddZdS)Timebase numerator denominatorrrN)rrreturncCs||_||_dS)Nr )selfrrr=/tmp/pip-unpacked-wheel-vx7f76es/torchvision/io/_video_opt.py__init__szTimebase.__init__)__name__ __module__ __qualname__int__annotations__ __slots__rrrrrr s  r c @sDeZdZeeeeeeeedZdddddddd gZd d d d Zd S) VideoMetaData) has_videovideo_timebasevideo_duration video_fps has_audioaudio_timebaseaudio_durationaudio_sample_raterrrrr r!r"r#N)rcCs@d|_tdd|_d|_d|_d|_tdd|_d|_d|_dS)NFrr g) rr rrrr r!r"r#)rrrrr9s  zVideoMetaData.__init__) rrrboolr floatrrrrrrrr#s&  r) pts_rangercCs@|d|dkrdkrr?r@rArBrCrDrrErFrGr5r!resultvframes _vframe_ptsr)r*r+r3r4r,r-r.inforrr_read_video_from_file{s8A  rO)r=rc Cs~tjj|dddddddddddddddddd}|\ }}}}}}}} } } t|||| | | } |}|}||| fS)z Decode all video- and audio frames in the video. Only pts (presentation timestamp) is returned. The actual frame pixel data is not copied. Thus, it is much faster than read_video(...) rr r6)rHrIr rJr2numpytolist) r=rK_vframes vframe_ptsr)r*r+_aframesr4r,r-r.rNrrr _read_video_timestamps_from_files4  rUc Cs4tjj|}|\}}}}}}t||||||}|S)zO Probe a video file and return VideoMetaData with info about the video )rHrIr Zprobe_video_from_filer2) r=rKr)r*r+r,r-r.rNrrr_probe_video_from_filesrV) video_datar>r?r@rArBrCrDvideo_timebase_numeratorvideo_timebase_denominatorrErFrGr5audio_timebase_numeratoraudio_timebase_denominatorrcCst|t| t|tjs,tj|tjd}tjj||d||||||d|d|| | | | | d| d||}|\ }}}}}}}}}}| dkrt ||| }||fS)a Reads a video from memory, returning both the video frames as well as the audio frames This function is torchscriptable. Args: video_data (data type could be 1) torch.Tensor, dtype=torch.int8 or 2) python bytes): compressed video content stored in either 1) torch.Tensor 2) python bytes seek_frame_margin (double, optional): seeking frame in the stream is imprecise. Thus, when video_start_pts is specified, we seek the pts earlier by seek_frame_margin seconds read_video_stream (int, optional): whether read video stream. If yes, set to 1. Otherwise, 0 video_width/video_height/video_min_dimension/video_max_dimension (int): together decide the size of decoded frames: - When video_width = 0, video_height = 0, video_min_dimension = 0, and video_max_dimension = 0, keep the original frame resolution - When video_width = 0, video_height = 0, video_min_dimension != 0, and video_max_dimension = 0, keep the aspect ratio and resize the frame so that shorter edge size is video_min_dimension - When video_width = 0, video_height = 0, video_min_dimension = 0, and video_max_dimension != 0, keep the aspect ratio and resize the frame so that longer edge size is video_max_dimension - When video_width = 0, video_height = 0, video_min_dimension != 0, and video_max_dimension != 0, resize the frame so that shorter edge size is video_min_dimension, and longer edge size is video_max_dimension. The aspect ratio may not be preserved - When video_width = 0, video_height != 0, video_min_dimension = 0, and video_max_dimension = 0, keep the aspect ratio and resize the frame so that frame video_height is $video_height - When video_width != 0, video_height == 0, video_min_dimension = 0, and video_max_dimension = 0, keep the aspect ratio and resize the frame so that frame video_width is $video_width - When video_width != 0, video_height != 0, video_min_dimension = 0, and video_max_dimension = 0, resize the frame so that frame video_width and video_height are set to $video_width and $video_height, respectively video_pts_range (list(int), optional): the start and end presentation timestamp of video stream video_timebase_numerator / video_timebase_denominator (float, optional): a rational number which denotes timebase in video stream read_audio_stream (int, optional): whether read audio stream. If yes, set to 1. Otherwise, 0 audio_samples (int, optional): audio sampling rate audio_channels (int optional): audio audio_channels audio_pts_range (list(int), optional): the start and end presentation timestamp of audio stream audio_timebase_numerator / audio_timebase_denominator (float, optional): a rational number which denotes time base in audio stream Returns: vframes (Tensor[T, H, W, C]): the `T` video frames aframes (Tensor[L, K]): the audio frames, where `L` is the number of points and `K` is the number of channels Zdtyperr ) r( isinstancerHTensor frombufferuint8rIr read_video_from_memoryr/r:)rWr>r?r@rArBrCrDrXrYrErFrGr5rZr[rKrLrMr)r*r+r3r4r,r-r.rrr_read_video_from_memory s:F   rb)rWrc Cst|tjstj|tjd}tjj|dddddddddddddddddd}|\ }}}}}}}} } } t|||| | | } | }| }||| fS)z Decode all frames in the video. Only pts (presentation timestamp) is returned. The actual frame pixel data is not copied. Thus, read_video_timestamps(...) is much faster than read_video(...) r\rr r6) r]rHr^r_r`rIr rar2rPrQ) rWrKrRrSr)r*r+rTr4r,r-r.rNrrr"_read_video_timestamps_from_memoryts8   rcc CsPt|tjstj|tjd}tjj|}|\}}}}}}t||||||}|S)zy Probe a video in memory and return VideoMetaData with info about the video This function is torchscriptable r\) r]rHr^r_r`rIr Zprobe_video_from_memoryr2) rWrKr)r*r+r,r-r.rNrrr_probe_video_from_memorys  rdpts)r= start_ptsend_ptspts_unitrc sdkrtddkr"tdt|}|j}|j}fdd}d}t} |rlt|jj |jj } || }d} t} |rt|j j |j j } || } t |d|| d| | d\} } }i}|r|j |d <|r|j|d <| | |fS) NinfremThe pts_unit 'pts' gives wrong results and will be removed in a follow-up version. Please use pts_unit 'sec'.cs`}}dkrHttd|}|tdkrHttd|}|tdkrXd}||fS)Nsecr rir6)rmathfloorr%ceil)Z time_baseZ start_offsetZ end_offsetrgrhrfrrget_ptss  z_read_video..get_ptsr<T)r?rDrrEr5r!rZ audio_fps)r%warningswarnrVrr default_timebaserrrrr!rOrr#)r=rfrgrhrNrr rprDrr5r!rLr3_inforror _read_videosF    ru)r=rhrcsd|dkrtdt|\}}}|dkrLt|jj|jjfdd|D}|jrX|jnd}||fS)Nrerjrkcsg|] }|qSrr).0xZvideo_time_baserr sz*_read_video_timestamps..) rqrrrUrrrrrr)r=rhre_rNrrrxr_read_video_timestampssr{)r;r rrrrr<rr r rrr<rr )rNre)re)#rlrqZ fractionsrtypingrrrrrrH extensionr Z_HAS_VIDEO_OPT ImportErrorOSErrorrsr rrr(r^r2r:strr%r$rOrUrVrbrcrdrur{rrrrs     !  !     a"#    l (   >