Hi Patrick,

This looks good to me, except one detail:

Since the ROI is given in binned coordinates, we won't be able to deal
with bins that start at coordinates that aren't a multiple of of the
binning factor. Is there a compelling reason why the ROI should be
given in binned coordinates rather than in unbinned coordinates?

I also wanted to comment on the scaling while rectifying case that you
decided to reject. I think that people who want to do this still can
by rectifying to a different name space, and publishing a new
camera_info. In fact, I sometimes wonder whether it would make sense
to always publish a new camera_info after changing an image, rather
than relying on this structure that contains a mix of pre- and
post-rectification data. But since I'm not a consumer of this data, I
don't really have a compelling argument to make.

Cheers,
Blaise

On Fri, Oct 15, 2010 at 5:30 PM, Patrick Mihelich
<mihelich@willowgarage.com> wrote:
> NOTE: for discussion, please use ros-developers list
>
> The REP, in text and slightly-more-readable html format, is attached.
>
> Also note that time to discuss this REP is unfortunately rather tight. Since
> sensor_msgs/CameraInfo lives in the common_msgs stack, changes to the
> message definition must be finalized by Oct 22, only one week from today. We
> can quibble over related issues such as image_geometry API changes after
> that point, but I want to reach a consensus on the overall design and
> specific message changes as quickly as possible. If you have a stake in
> CameraInfo, please take a look as soon as you can.
>
> Cheers,
> Patrick
>
> Abstract
> ========
>
> This REP adds support for binning and alternative distortion models to
> sensors_msgs/CameraInfo.  It clarifies the correct interpretation of
> region of interest (ROI), including how it interacts with binning and
> distortion.
>
> In this REP, sensor_msgs/CameraInfo acquires binning_x and binning_y
> fields.  The distortion parameter array D is made variable-length, its
> meaning specified by an enum.  This REP clarifies ROI to mean the raw
> image region as captured by the camera and introduces a distinction
> between raw ROI and rectified ROI.  It adds appropriate new methods to
> the camera model classes in image_geometry, the recommended API for
> using CameraInfo.
>
> _______________________________________________
> ros-developers mailing list
> ros-developers@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-developers
>
>