[FFmpeg-devel] [PATCH 1/1] This change adds an encoder for Camera metadata motion. This is a type of sensor data associated with video, such as GPS, acceleration, gyro, and camera orientation. It does not encode video itself, but rather, this metadata.

[Hendrik Leppkes]

On Sat, Jul 8, 2017 at 12:24 AM, Louis O'Bryan
<louiso-at-google.com at ffmpeg.org> wrote:
>
> The use case is to write data to a new stream in the mp4 container. The
> encoder isn't changing the data. This data would reside in the same mp4
> container as video and audio streams. Are you suggesting there is a way I
> can accomplish that task without creating this encoder (and if so, how)?

You don't need an encoder to create AVPackets, you can just do that
yourself, even more so if the encoder apparently doesn't do any
encoding.
So why does it need encoding in the first place? Why can't you just
pass the finished data in an AVPacket to the mov muxer? Thats how data
streams are supposed to work.

>
>> +static av_cold int encode_init(AVCodecContext *avctx) {
>> > +    // Use dummy values for the height and width.
>> > +    avctx->width = DUMMY_ENCODER_SIZE;
>> > +    avctx->height = DUMMY_ENCODER_SIZE;
>> > +    avctx->max_pixels = DUMMY_ENCODER_SIZE;
>> What? This makes no sense.
>
> Using avcodec_encode_video2() seems to require that the width and height be
> nonzero. What is the recommended way to avoid that?
>

Clearly you should not use a video encoder to encode something thats
not a video. This entire approach is not acceptable.