Re: V4L2_PIX_FMT_RAW

[Thierry Merle <thierry.merle@free.fr>]

Hi Thomas ,
Thomas Kaiser a écrit :
> Laurent Pinchart wrote:
>> On Thursday 21 February 2008, Thomas Kaiser wrote:
>>> 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!
>>
>> There are historical reasons. Those drivers should be fixed to remove
>> decoding from kernelspace. Obviously a new userspace component will
>> be needed to handle decoding and conversion, otherwise applications
>> will break. No consensus exists today regarding what form that
>> component should take.
>
> Yes, but all this transformation which is done in kernel space can be
> done in user space. But it looks like that nobody is interested to
> move this to user space (expect you) ;-)
> And I think it should not be that hard to introduce a user space
> component to handle this. When the user space app programmers are
> willing to do so!
>
Well, of course you can participate to the v4l2 library mailing-list.
We started a userspace daemon that would do the frame decompression in
userspace and give back uncompressed frames to the application via a
virtual device driver.
You will find more information on here:
http://www.linuxtv.org/v4lwiki/index.php/V4L2UserspaceLibrary
http://www.linuxtv.org/cgi-bin/mailman/listinfo/v4l2-library
I maintain the current code here:
http://linuxtv.org/hg/~tmerle/v4l2_extension

For now, the virtual driver is here, the userspace daemon is here, the
virtual driver <-> daemon command passing is bound to be here thanks to
Jiri Slaby.
Now, we need to implement the frame decompression. I should do that with
the usbvision driver that includes a decompression algorithm in
kernelspace but we can extend it to any v4l2 device driver.
Feel free to join, comment and ask for precisions!

>>
>>>> ...
>>>>
>>>> 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
>>
>> Best regards,
>>
>> Laurent Pinchart
>
> Best Regards,
>
> Thomas
>
Cheers,
Thierry

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