On Wed, 15 Jun 2011, Alexey Fisher wrote:
Am Dienstag, den 14.06.2011, 23:31 -0700 schrieb Carl Michal:
On Tue, 14 Jun 2011, Alexey Fisher wrote:
Am Montag, den 13.06.2011, 22:48 -0700 schrieb Carl Michal:
Hi,
I think you nailed it. Every frame looks perfect now. The trace shows a few
of these:
Jun 13 09:24:24 uvcvideo: Dropping payload (error bit set)
but I don't see corrupt frames any more in either MJPG or YUYV (at 640x480
anyway) - in MJPG all the frames have the right size.
There is a some weirdness with frame rates depending on the exposure setting:
1) Exposure, auto gives 4 options: auto priority mode, manual mode, shutter
priority mode, and aperture priority mode. Auto and shutter don't seem to be
settable (errors from guvcview when chosen). There is also an "Exposure, auto
priority" checkbox.
Frame rates drop dramatically in manual mode (to 10-15fps from 30).
But I can't really complain at this point - the corrupt frames are gone.
Will that quirk be added to the driver (usb id is: 0408:2fb1)?
Thanks,
Hi,
it seems like I am much better off by fully disabling FID (with your
patch) than before. With the patch, YUYV frames are _always_ the right
size. There are still some problems:
1) corrupt frames - with part of the image missing or the image displaced.
Sometimes (but definitely not always) these occur at the same time as a
trace message saying the error bit is set.
2) sometimes the camera just won't start. when guvcview (or luvcview) is
started, no frames come back from the camera. There is a light next to
the camera that comes on to indicate it should be active, but no frames
arrive. There seems to be a fairly strong correlation with using luvcview
(which from the traces seems to use some different mechanism to get frames from
the driver from guvcview. guvcview polls, luvcview doesn't). Sometimes
just restarting guvcview several times will work and the camera eventually
starts. Sometimes just changing resolution or frame rates succeeds in
starting the camera. I haven't found anything reproducible. I do not
think this is related to your patch, as it did happen once before your
patch was applied. Unloading and reloading the uvcvideo and ehci_hcd
modules does not consistently solve it. guvcview just lists:
Could not grab image (select timeout): Resource temporarily unavailable
and the trace shows guvcview polling, but nothing else going on.
I have tried adding the other quirks to the FID quirk, but haven't seen
any improvement with any others.
Thanks for you help -
Carl
Webcam returns error in the middle of some frame, theoretically we
should drop complete frame. But current uvcvideo just gather data and
assume the cam will resend previous parts to complete the frame.
Try attached patch additionally to my previous one.
--
Regards,
Alexey
Hi,
its very hard to say if this helps or not. There are still corrupt
frames, and some seem to occur at about the same time as the error bit
trace messages, but some don't show anything unusual in the traces that
I've noticed yet.
Since all the uncompressed frames were the right size (even ones where the
error bit was set somewhere) those frames are at least complete.
Is there some convenient way to capture just those frames with the error
bit set?
Thanks,
Carl
you can try this command:
gst-launch-0.10 -v v4l2src ! video/x-raw-yuv,width=320 ! identity ! jpegenc !
multifilesink location=tmp-%05d.jpg
it will produce for each frame one jpeg file. Watch out, it will produce
lots of files.
Attached patch is replacement for the last one. You do not need setting
trace option, it will print all we need.
--
Regards,
Alexey
Hi,
we're learning a little here. If I set nodrop=1, I get lots of frames
that are too short (gst complains that they are fewer bytes than expected)
but those coincide with the status < 0 test in uvc_video_decode_isoc -
they do not correspond to those with the error bit set.
The buf->error=1 in uvc_video_decode_start=1 is unnecessary - harmful
even, since it means dropping frames that are in fact ok.
I have captured some bad frames though - one that has the colors badly
screwed up - where everything is a bright green, and another where the
image appears to be shifted about a quarter of a frame to one side - these
were collected without the nodrop=1 parameter, so they are not caught as
corrupt by any of the current tests. They are delivered as occuring with
the full expected frame size. Are there other flags we should be checking
for? (hmm, like maybe an FID at the wrong time?)
Carl
_______________________________________________
Linux-uvc-devel mailing list
Linux-uvc-devel@lists.berlios.de
https://lists.berlios.de/mailman/listinfo/linux-uvc-devel
