On 14/04/14 17:27, Roman Savchenko wrote:
> Ill provide some code samples (it's not purely correct), what i mean:
>
> static int create_side_data(H264Picture *pic)
>
> {
>
> int i, motion_val_size = 0;
>
> AVFrameSideData *mb_type_side_data = NULL;
>
> AVFrameSideData *motion_val_side_data = NULL;
>
>
> mb_type_side_data = av_frame_new_side_data(&pic->f,
> AV_FRAME_DATA_MB_TYPE, sizeof(AVBufferRef));
>
> if (!mb_type_side_data)
>
> goto fail;
>
>
> memcpy(mb_type_side_data->data, pic->mb_type_buf, sizeof(AVBufferRef));
>
>
> motion_val_size = sizeof(pic->motion_val_buf) /
> sizeof(pic->motion_val_buf[0]);
>
> motion_val_side_data = av_frame_new_side_data(&pic->f,
> AV_FRAME_DATA_MOTION_VAL,
>
> motion_val_size * sizeof(AVBufferRef));
>
>
> if (!motion_val_side_data)
>
> goto fail;
>
>
> for (i = 0; i < motion_val_size; ++i) {
>
> AVBufferRef *ref =
> (AVBufferRef*)(&(motion_val_side_data->data[i *
> sizeof(AVBufferRef)]));
>
> memcpy(ref, pic->motion_val_buf[i], sizeof(AVBufferRef));
>
> }
>
>
> return 0;
>
>
> fail:
>
> if (mb_type_side_data)
>
> av_frame_remove_side_data(&pic->f, AV_FRAME_DATA_MB_TYPE);
>
> if (motion_val_side_data)
>
> av_frame_remove_side_data(&pic->f, AV_FRAME_DATA_MOTION_VAL);
>
>
> return AVERROR(ENOMEM);
>
> }
>
>
>
>
> 2014-04-14 16:29 GMT+03:00 Roman Savchenko <[email protected]>:
>
>> Hi All,
>>
>> a little bit intro: some times ago I asked at libav-user mailing list
>> about how can I use mb_type and motion_val from H264Picture. They are used
>> in some detection algo that i wrote (its called "Detection in compressed
>> domain"). Luca Barbato answered that it can be a part of side data. But no
>> future discussion are present. And I decided "to patch" by myself.
>>
>> But unfortunately I met a problem, because I don't fully understand
>> architecture idea of some things. And, it'll be very nice, if u'll help me
>> with some advice.
I was waiting either Anton or Vittorio to chip in since they used that
part of Libav the most.
>>
>> For first I add AV_FRAME_DATA_MB_TYPE and AV_FRAME_DATA_MOTION_VAL in
>> AVFrameSideData enum.
>>
>> For second i try to decide what type (or rather copy) in side data I
>> should use. Because If I'll use AVBufferRef I must (should I?) to ref with
>> av_buffer_ref. But it will provide a allocation for AVBufferRef and
>> AVFrameSideData will provide allocation for AVBufferRef too. It'll not be
>> a problem if I will copy by value.
Probably would make sense rethinking side data since the amount of
information exported isn't small in this specific case.
>> For third, its pretty the same as second, what type of copy should I use
>> to copy H264Picture.motion_val. If it will be a deep copy should I expand
>> memory to one line or simply provide AV_FRAME_DATA_MOTION_VAL0,
>> AV_FRAME_DATA_MOTION_VAL1?
>>
>> For fourth, and last, can I use expanded_data rather side_data?
Might be interesting in a way but I'm afraid overloading the meaning of
those pointers would be quite annoying.
>> Where should I store knowledge about indexes?
An option would be baking a struct containing all the information and
have a custom destructor to unref properly what's there.
lu
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
