On Thursday 10 January 2002 09:23 am, Alan Cox wrote: > Don't scale it - you damage the data as well. You have a good point though > and people are working on updating the API for video in the kernel. The > pixel shape one is a very relevant point. Is this the V4L2 stuff I read about on thedirks.org/v4l2 ? I'd like to work with the group doing the new API if only to ensure that the issues with this cam have visibility. > The reason we complain to people about format conversions is that most > don't need them. If your camera is a YUYV like format convert to a > supported YUV one, if its RGB based then RGB888 is great. YUV->RGB in > kernel is not nice generally. Unfortunately it requires the same number of fixed and floating point ops to convert this format to a proper YUV as RGB because its color wheel is rotated and the UV information is packed in an unusual manner (I'm very thankful somebody else managed to reverse engineer the colorspace encoding). > For the moment add a private ioctl, but this is also relevant to fixing the > API Will do... but let me know where the new API work is being done so I can get the camera supported on it and make sure that things like non-square pixels and application-configurable shutter speeds are addressed as video capabilities. -Joe
