/// <summary>
/// transcode audio
/// </summary>
/// <param name="input">input audio file</param>
/// <param name="output">output audio file</param>
/// <param name="outChannels">output audio file channels</param>
/// <param name="outSampleRate">output audio file sample rate</param>
public AudioTranscode(string input, string output, int outChannels = 2, int outSampleRate = 44100)
{
using (MediaWriter writer = new MediaWriter(output))
using (MediaReader reader = new MediaReader(input))
{
int audioIndex = reader.First(_ => _.Codec.Type == AVMediaType.AVMEDIA_TYPE_AUDIO).Index;
writer.AddStream(MediaEncoder.CreateAudioEncode(writer.Format, outChannels, outSampleRate));
writer.Initialize();
AudioFrame dst = AudioFrame.CreateFrameByCodec(writer[0].Codec);
SampleConverter converter = new SampleConverter(dst);
long pts = 0;
foreach (var packet in reader.ReadPacket())
{
foreach (var srcframe in reader[audioIndex].ReadFrame(packet))
{
foreach (var dstframe in converter.Convert(srcframe))
{
pts += dstframe.AVFrame.nb_samples;
dstframe.Pts = pts; // audio's pts is total samples, pts can only increase.
foreach (var outpacket in writer[0].WriteFrame(dstframe))
{
writer.WritePacket(outpacket);
}
}
}
}
writer.FlushMuxer();
}
}
/// <summary>
/// decode video to image
/// filter graph:
/// ┌──────┐ ┌──────┐ ┌─────┐ ┌──────────┐ ┌──────┐
/// │input0│---->│buffer│---->│scale│---->│buffersink│---->│output│
/// └──────┘ └──────┘ └─────┘ └──────────┘ └──────┘
/// </summary>
/// <param name="inputFile">input video file</param>
/// <param name="outDirectory">folder for output image files</param>
/// <param name="scaleOptions">scale options <see cref="http://ffmpeg.org/ffmpeg-filters.html#scale-1"/></param>
public DecodeVideoWithCustomCodecScaledToMat(string inputFile, string outDirectory, string scaleOptions = "512:288")
{
using (MediaReader reader = new MediaReader(inputFile, null, null))
{
var videoIndex = reader.First(_ => _.Codec.AVCodecContext.codec_type == AVMediaType.AVMEDIA_TYPE_VIDEO).Index;
unsafe
{
// relpace the default vide decode
// !!! IMPORTANT NOTE: This sample won't work, if you haven't downloaded ffmpeg (GPL license, as it is more complete), and you don't have NVIDIA hardware (CUDA) !!!
reader[videoIndex].Codec = MediaDecode.CreateDecode("h264_cuvid", _ => ffmpeg.avcodec_parameters_to_context(_, reader[videoIndex].Stream.codecpar));
}
int height = reader[videoIndex].Codec.AVCodecContext.height;
int width = reader[videoIndex].Codec.AVCodecContext.width;
int format = (int)reader[videoIndex].Codec.AVCodecContext.pix_fmt;
AVRational time_base = reader[videoIndex].TimeBase;
AVRational sample_aspect_ratio = reader[videoIndex].Codec.AVCodecContext.sample_aspect_ratio;
/* We are moving the packet to CUDA to perform the scaling.
* We can then:
* - remove hwdownload and format to leave it in CUDA, and forward the pointer to any other function, or write the frame to the output video
* - convert it to MAT whereas converting speed depends on the size of the scaled frame.
*/
MediaFilterGraph filterGraph = new MediaFilterGraph();
filterGraph.AddVideoSrcFilter(new MediaFilter(MediaFilter.VideoSources.Buffer), width, height, (AVPixelFormat)format, time_base, sample_aspect_ratio)
.LinkTo(0, filterGraph.AddFilter(new MediaFilter("scale"), scaleOptions))
.LinkTo(0, filterGraph.AddVideoSinkFilter(new MediaFilter(MediaFilter.VideoSinks.Buffersink)));
filterGraph.Initialize();
var sw = Stopwatch.StartNew();
foreach (var packet in reader.ReadPacket())
{
foreach (var frame in reader[videoIndex].ReadFrame(packet))
{
filterGraph.Inputs.First().WriteFrame(frame);
foreach (var filterFrame in filterGraph.Outputs.First().ReadFrame())
{
using (var image = filterFrame.ToMat())
{
image.Save(Path.Combine(Directory.CreateDirectory(outDirectory).FullName, $"{DateTime.Now.Ticks}.jpg"));
}
}
}
}
Console.WriteLine($"Converting to MAT [ processed in {sw.Elapsed.TotalMilliseconds:0} ms ]");
}
}