Why do we need a buffer while converting AVFrame from one format to another?

c, ffmpeg

Solution

I think that what puzzles you is that there seem to be two allocations for AVFrame.

The first, done with `avcodec_alloc_frame()`, allocates the space for a generic frame and its metadata. At this point the memory required to hold the frame proper is still unknown.

You then populate that frame from another source, and it is then that you specify how much memory you need by passing `width`, `height` and color depth:

numBytes=avpicture_get_size(PIX_FMT_RGB24, pCodecCtx->width, pCodecCtx->height);

At this point the frame and its content are two separate objects (an AVFrame and its buffer). You put them together with this code, which is not actually a conversion at all:

avpicture_fill((AVPicture *)pFrameRGB, buffer, PIX_FMT_RGB24,
pCodecCtx->width, pCodecCtx->height);

What the code above does is to "tell" `pFrameRGB`: " you are a RGB-24 frame, this wide, this tall, and the memory you need is in 'buffer' ".

Then and only then you can do whatever you want with `pFrameRGB`. Otherwise, you try to paint on a frame without the canvas, and the paint splashes down -- you get a core dump.

Once you have the frame (AVFrame) and the canvas (the buffer), you can use it:

// Read frames and save first five frames to disk
i=0;
while(av_read_frame(pFormatCtx, &packet)>=0) {
    // Is this a packet from the video stream?
    if(packet.stream_index==videoStream) {
      // Decode video frame
      avcodec_decode_video2(pCodecCtx, pFrame, &frameFinished,
&packet);

The above code extracts a video frame and decodes it into `pFrame` (which is native format). We could save `pFrame` to disk at this stage. We would not need `buffer`, and we could then not use `pFrameRGB`.

Instead we convert the frame to RGB-24 using `sws_scale()`.

To convert a frame into another format, we copy the source to a different destination. This is both because the destination frame could be bigger than what can be accommodated by the source frame, and because some conversion algorithms need to operate on larger areas of the untransformed source, so it would be awkward to transmogrify the source in-place. Also, the source frame is handled by the library and might conceivably not be safe to write to.

Update (comments)

What does the `data[]` of pFrame/pFrameRGB point to: initially, nothing. They are NULL, and that is why using a noninitialized AVframe results in a core dump. You initialize them (and `linesize[]` etc.) using `avpicture_fill` (that fits in an empty buffer, plus image format and size information) or one of the decode functions (which do the same).

Why does pFrame not require memory allocation: good question. The answer is in the used function's prototype and layout, where the picture parameter is described thus:

The AVFrame in which the decoded video frame will be stored. Use avcodec_alloc_frame to get an AVFrame, the codec will allocate memory for the actual bitmap. with default get/release_buffer(), the decoder frees/reuses the bitmap as it sees fit. with overridden get/release_buffer() (needs CODEC_CAP_DR1) the user decides into what buffer the decoder decodes and the decoder tells the user once it does not need the data anymore, the user app can at this point free/reuse/keep the memory as it sees fit.

Problem

I am referring to this source code . The code snippets provided here are from lines (114-138) in the code . This is using the ffmpeg library . Can anyone explain why is the following code required in the program ? ``` // Determine required buffer size and allocate buffer numBytes=avpicture_get_size(PIX_FMT_RGB24, pCodecCtx->width, pCodecCtx->height); buffer=(uint8_t *)av_malloc(numBytes*sizeof(uint8_t)); ``` In a sense I understand that the following function is associating the destination frame to the buffer . But what is the necessity ? ``` avpicture_fill((AVPicture *)pFrameRGB, buffer, PIX_FMT_RGB24, pCodecCtx->width, pCodecCtx->height); ``` PS : I tried removing the buffer and compiling the program . It got compiled . But it is showing the following run time error . [swscaler @ 0xa06d0a0] bad dst image pointers Segmentation fault (core dumped)

Original source