[Ffmpeg-devel] de/serializing AVFrames

[Tilman Kranz]

Hello,

Michael Niedermayer wrote:
 > Tilman Kranz wrote:
>> [...] I would like this cache to reside on disk if neccessary [...]
> you do know about AVCodecContext.get/release_buffer() and they arent
> doing what you want right?

No, I not really know them - thanks for pointing that out!

My main routine is (simplified):

--- pseudocode start ---
AVFrame *pFrame = avcodec_alloc_frame()
cache_t *cache = NULL;

// ... prepare decoder context with video stream ...

while
(decode_next_frame(formatCtx,codecCtx,videoStream,pFrame))

    cache = cache_append
       (cache, cache_new
          (pFrame,
           codecCtx->pix_fmt,
           codecCtx->width,
           codecCtx->height));

// ... encode in different order ...
// ... done.
--- pseudocode stop ---

A cache entry is initialized as follows (simplified):

--- pseudocode start ---
cache_t *cache_new
    (AVFrame *frame, enum PixelFormat pix_fmt, int width, int height)
{
    cache_t *result = new cache_t;

    result->frame = avcodec_alloc_frame();
    result->buffer = new uint8_t[avpicture_get_size(...)];

    // ... keep pix_fmt, width, height ...

    avpicture_fill(result->frame,result->buffer,...);
    img_copy((AVPicture *)result->frame,(AVPicture *)frame, ...);

    result->on_disk = false;

    return result;
}
--- pseudocode stop ---

In "cache_new()" it saved me the "img_copy()" if I could reuse the
buffer from the AVFrame that is in codec context and let the codec
allocate a new one. Is this what "get/release_buffer()" do? If I
re-get a buffer, do I have to call "avpicture_fill()" afterwards
to assign offsets and sizes?

 > [...]
>>Also, if there is a higher-level way to do it or it is
>>likely to break in the future please let me know.
> no highlevel way, unlikely to break once it works ...

Cool, thanks again.

Greetings, Tilman.