No subject

[bogus at does.not.exist.com]

> +  const uint8_t* const buf = avpkt->data;
> +  const int buf_size = avpkt->size;
> +  vp8dctx_t* const ctx = avctx->priv_data;
> +  AVFrame* const picture = data;

These const do nothing but making code changes harder, and we don't
use them this way anywhere else. Also in FFmpeg the * goes with the variable name,
not the type.
I also fail to see the point of the extra buf and buf_size variables,
they seem to be used only once.

> +  if(vpx_codec_decode(&ctx->decoder,buf,buf_size,NULL,0)!=VPX_CODEC_OK) {

To handle codec delay, buf == NULL and buf_size == 0 is valid in FFmpeg,
is using it this way also part of the VPX API and has the same meaning?
Otherwise it would be better to handle it specially.
Also, does the decoder somehow tell how much of the data it actually used?
On the cosmetic side, this needs a few more spaces to match the general coding style.

> +    picture->data[0] = img->planes[0];
> +    picture->data[1] = img->planes[1];
> +    picture->data[2] = img->planes[2];
> +    picture->data[3] = img->planes[3];
> +    picture->linesize[0] = img->stride[0];
> +    picture->linesize[1] = img->stride[1];
> +    picture->linesize[2] = img->stride[2];
> +    picture->linesize[3] = img->stride[3];

And while as someone else mentioned you might not have to copy the [3] entries, 
I think they should initialized.