Recherche avancée

Sur d’autres sites (11937)

• Book Author not added to metadata

  1er décembre 2013, par Grandt

  Book Author not added to metadata

• Why does ffmpeg parses an entire file beforehand

  10 juillet 2023, par Dmitrii Dunaev

  When calling avformat_open_input with AVIOContext set it seems that the library parses an entire file. I set both read and seek functions on the context. The video file is a fragmented mp4 with a lot of moof boxes. The file is located on a remote server. Each seek call is transformed into an http request. The library issues one seek for each moof. Is there any way to prevent avformat_open_input from parsing the entire file beforehand ? I would like to open the file and decode one frame at a specific time offset. Any suggestions ?

  


  Here is the snippet of GO code

  


  func (d *videoDecoder) init() error {
    avio_ctx_buffer := (*Cuint8_t)(lib.av_mallocz(avio_buffer_size))

    d.format_ctx = lib.avformat_alloc_context()
    d.avio_ctx = lib.avio_alloc_context(avio_ctx_buffer, Cint(avio_buffer_size), 0, uintptr(unsafe.Pointer(&d.ReadSeeker)), (Crw_packet_callback)(C.read_packet), nil, (Cseek_packet_callback)(C.seek_packet))
    d.format_ctx.pb = d.avio_ctx

    if ret := lib.avformat_open_input(&d.format_ctx, nil, nil, &d.options); ret < 0 {
        return fmt.Errorf("avformat open input: %w", convertRetCode(ret))
    }

    return nil
}

func read_packet(opaque unsafe.Pointer, buffer *Cuint8_t, size Cint) Cint {
    r := *(*io.ReadSeeker)(opaque)

    gobuf := (*[1 << 30]byte)(unsafe.Pointer(buffer))[:size:size]
    fmt.Printf("REQUESTED TO READ: %v\n", size)

    n, err := r.Read(gobuf)
    if err != nil {
        if errors.Is(err, io.EOF) {
            return CAVERROR_EOF
        }
        fmt.Printf("error reading from reader: %v\n", err)
    }

    return Cint(n)
}

//export seek_packet
func seek_packet(opaque unsafe.Pointer, offset Cint64_t, whence Cint) Cint64_t {
    r := *(*io.ReadSeeker)(opaque)

    fmt.Printf("REQUESTED TO SEEK: whence: %v offset: %v\n", int(whence), int64(offset))

    if whence == CAVSEEK_SIZE || whence == CAVSEEK_FORCE{
        return -1
    }

    n, err := r.Seek(int64(offset), int(whence))
    if err != nil {
        if errors.Is(err, io.EOF) {
            return -1
        }
        fmt.Printf("error seeking: %v\n", err)
    }

    return Cint64_t(n)
}


  


  And here is the output (truncated). It seeks to every single moof box.

  


  REQUESTED TO READ: 4196
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 4196
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 4196
REQUESTED TO SEEK: whence: 0 offset: 615080
REQUESTED TO READ: 4196
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 619276
REQUESTED TO SEEK: whence: 0 offset: 1232664
REQUESTED TO READ: 4196
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 65536 offset: 0
REQUESTED TO SEEK: whence: 2 offset: -1
REQUESTED TO SEEK: whence: 0 offset: 1236860
REQUESTED TO SEEK: whence: 0 offset: 1845238
REQUESTED TO READ: 4196
...


  


• dnn : put DNNModel.set_input and DNNModule.execute_model together

  10 septembre 2020, par Guo, Yejun

  dnn : put DNNModel.set_input and DNNModule.execute_model together
  suppose we have a detect and classify filter in the future, the
  
  detect filter generates some bounding boxes (BBox) as AVFrame sidedata,
  
  and the classify filter executes DNN model for each BBox. For each
  
  BBox, we need to crop the AVFrame, copy data to DNN model input and do
  
  the model execution. So we have to save the in_frame at DNNModel.set_input
  
  and use it at DNNModule.execute_model, such saving is not feasible
  
  when we support async execute_model.
  This patch sets the in_frame as execution_model parameter, and so
  
  all the information are put together within the same function for
  
  each inference. It also makes easy to support BBox async inference.
  

  • [DH] libavfilter/dnn/dnn_backend_native.c
  • [DH] libavfilter/dnn/dnn_backend_native.h
  • [DH] libavfilter/dnn/dnn_backend_openvino.c
  • [DH] libavfilter/dnn/dnn_backend_openvino.h
  • [DH] libavfilter/dnn/dnn_backend_tf.c
  • [DH] libavfilter/dnn/dnn_backend_tf.h
  • [DH] libavfilter/dnn_interface.h
  • [DH] libavfilter/vf_derain.c
  • [DH] libavfilter/vf_dnn_processing.c
  • [DH] libavfilter/vf_sr.c