Re: V4L2_PIX_FMT_RAW

[Thomas Kaiser <linux-dvb@kaiser-linux.li>]

H. Willstrand wrote:
> On Thu, Feb 21, 2008 at 7:55 PM, Thomas Kaiser
> <linux-dvb@kaiser-linux.li> wrote:
>> H. Willstrand wrote:
>>  > On Thu, Feb 21, 2008 at 1:00 PM, Thomas Kaiser
>>  > <linux-dvb@kaiser-linux.li> wrote:
>>  >> H. Willstrand wrote:
>>  >>  > On Thu, Feb 21, 2008 at 2:20 AM, Daniel Glöckner <daniel-gl@gmx.net> wrote:
>>  >>  >> On Thu, Feb 21, 2008 at 01:02:39AM +0100, H. Willstrand wrote:
>>  >>  >>  > What's the problem with having a name of the formalized data in the
>>  >>  >>  > video stream? ie raw do not mean undefined.
>>  >>  >>
>>  >>  >>  I thought you wanted to avoid having to define V4L2_PIX_FMT_x for an
>>  >>  >>  exploding number of proprietary formats that are quite similar but still
>>  >>  >>  incompatible. It makes sense for formats that are used by more than one
>>  >>  >>  driver.
>>  >>  >
>>  >>  > Correct, the number of unique pixel formats should be kept down.
>>  >>  > Again, comparing with digital cameras there are >200 proprietary
>>  >>  > formats and there is a "clean-up" on-going where the "market" is
>>  >>  > aiming for a OpenRAW.
>>  >>  >
>>  >>  > However, by declaring a generic RAW format (which is then driver
>>  >>  > specific) doesn't help the user mode app developers. Calling a
>>  >>  > multitude of libraries to see if you get lucky might not be a good
>>  >>  > idea.
>>  >>  >
>>  >>  > Still, I'm suspectious about the definition "raw" used here.
>>  >>  > RAW should mean unprocessed image data:
>>  >>  > * no white balance adjustment
>>  >>  > * no color saturation adjustments
>>  >>  > * no contrast adjustments
>>  >>  > * no sharpness improvements
>>  >>  > * no compression with loss
>>  >>
>>  >>  Yes, raw means "as it is" no stripping, decoding  or removing of SOF headers are
>>  >>  done in the driver. May be V4L2_PIX_FMT_AII (AII -> As It Is) is the better name?
>>  >>
>>  >
>>  > I struggle with the probability to find several CCD's having similar
>>  > formats. There aren't so many manifactors of CCD's but they truelly
>>  > can generate divergeting formats. Worst case scenario means >200
>>  > V4L2_PIX_FMT_RAW_...
>>  >
>>  > I think RAW is a OK name, the question is if the subcomponents of the
>>  > RAW formats has similarities, if so they might be standardized.
>>  > Looking into different Sony CCD's it's clearly possible, but after the
>>  > CCD the data has to be buffered, packaged and transmitted which of
>>  > course can be done in several ways...
>>  >
>>  > Cheers,
>>  > Harri
>>  >
>>  >>  >
>>  >>  > So, by looking for similarities in the "raw" formats where available
>>  >>  > there should be a potential to consolidate them.
>>  >>  >
>>  >>  >>
>>  >>  >>  > I don't see how separate RAW ioctl's will add value to the V4l2 API,
>>  >>  >>  > it fits into the current API.
>>  >>  >>
>>  >>  >>  Yes, it does. Each driver having multiple raw formats just needs a
>>  >>  >>  private control id to select one.
>>  >>  >>
>>  >>  > I was more thinking about the VIDIOC_S_RAW stuff, a VIDIOC_S_FMT
>>  >>  > should do the job.
>>  >>  > I.e. I think there should be strong reasons to break V4L2 API behavior.
>>  >>  >
>>  >>  > Harri
>>
>>  Actually, in a webcam you have the image sensor and a usb bridge. Usually, the
>>  sensor capture a picture in Bayer pattern. This gets forwarded to the usb
>>  bridge. The usb bridge may or may not transfer the picture to an other format
>>  and/or compress it with a standard compression algo or a proprietary compression
>>  algo. The resulting data stream will be transmitted over the usb interface.
>>
> 
> Yes, the USB bridge buffers, packages and transmits.
> 
>>  I just would like to get this resulting stream to user space without
>>  manipulation/conversion/decoding of the stream in the kernel module.
>>
>>  That means we don't know what the format is in this data which comes trough the
>>  usb interface. That's way I call it raw.
>>
>>  At the moment with V4L2, I have to forward a stream to user space which is in a
>>  format v4l2 knows. That means I have sometimes to do heavy data processing in
>>  the kernel module to decode/convert the data from the usb stream to a known v4l2
>>  video format.
>>
> 
> Drivers should not do any decoding / converting, it's not allowed in
> kernel mode.
> But you are right, there are a number of V4L1 exceptions:
> AR M64278 (arv.c) converts YUV422 to YUV422P
> QuickCam (bw-qcam.c) converts RAW to a useful format :)
> CPiA (cpia.c) converts 420 to different RGB formats
> OmniVision (ov511.c) converts from YUV4:0:0
> PWC (V4L2) does decoding

You forgot gspca [1](support of 260 webcams at the moment) and here we even do 
jpeg decoding in kernel space to get the proper format for v4l1!

> ...
> 
> However, the Webcams provides only a limited set of formats and the
> "raw" are usually available. New drivers with proprietary "raw"
> formats should be added to videodev2.h

That means you agree with me?

> 
>>  That's way I want a official way to forward the untouched usb stream to user space!
>>
>>  How the user space application has to react on this stream is an other story, I
>>  think. But there will be some way to tell the usespace application what to do
>>  with this "unknown" stream, I am sure.
>>
>>  Thomas
>>
> 
> Cheers,
> Harri

Thomas


[1] http://mxhaard.free.fr/download.html

-- 
http://www.kaiser-linux.li

--
video4linux-list mailing list
Unsubscribe mailto:video4linux-list-request@redhat.com?subject=unsubscribe
https://www.redhat.com/mailman/listinfo/video4linux-list