[libav-api] Understanding and working with sws_scale
hydrogen18 at gmail.com
Wed Mar 20 00:03:42 CET 2013
On Tue, Mar 19, 2013 at 4:10 AM, Luca Barbato <lu_zero at gentoo.org> wrote:
> On 19/03/13 01:54, Eric Urban wrote:
> > I have reverse engineered what appears to be a proprietary protocol used
> > many Chinese designed and manufactured security DVRs. I'm able to extract
> > h.264 frames from a TCP Data Stream. Initially I was able just writing
> > the frames one after another to a file and using avconv to convert to
> > another video format. The video looks fine, so I am assuming I am doing
> > correctly
Great, would be possible integrate it as an avdevice or a protocol?
That should be possible. Once I have a better documented proof on concept
I'll post to one of the lists and see if it is suitable for integration
into libav. Otherwise my intent is to license it with a LGPL-compatible
> > Since then I've gotten to the point where I can use libav to decode the
> > packets. This of course results in an AVFrame. Right now I am trying to
> > figure out a way to write out the most current frame to a still image. My
> > current understanding of libav is that the pixel format is depending on
> > source data, in this case h.264. Either way, it appears it is accessible
> > from decoding context struct AVCodecContext::pix_fmt.
> > Apparently if I just wanted to re-encode this in a better supported
> > streaming format it would be easy. But to make these security cameras
> > with open source software like motion I really need to be able to produce
> > stills. My logic is that if I can convert the pixels into something like
> > RGBA, I can from there easily write to whatever format I want like ppm,
> > jpeg, etc. Apparently this is done with the function sws_scale. The
> > is easy enough to create with sws_getContext. But I really have no idea
> > what the parameters to sws_scale should be.
> you could have a look at output-example.c
I found the 'libavformat/output-example.c' . I will read through it to
try and understand it better. Thanks.
> > What are slices? What are strides?
> slices: vertical section of the image at hand, number of rows.
> strides: linesizes, how many bytes to read before you get to the next row.
> > Can someone give me an explanation of how sws_scale works? I've found
> > and pieces of information scattered across the internet but nothing
> > concrete. Also, I could not figure out how to search this list. Is that
> > possible?
> swscale is relatively simple and ugly.
> you get a context telling which kind of scaler you want and on which
> kind of picture geometry will work on (from wxh to w1xh1)
> then you feed the scaler with rows of that image, you can feed the whole
> picture or just group of rows.
> a row is described by its stride (or linesize), since for performance
> reason your picture dimension and the actual size of a line (or row) can
> be different.
> I hope it helps.
> libav-api mailing list
> libav-api at libav.org
More information about the libav-api