Hi Neall,
The RGBA/YCA conversion code is convoluted mostly because of the buffer
management that is required to support random access to scan lines in
the API with reasonable efficiency. For example, if you read a YCA
file, then RGBA scan line n in your frame buffer depends on YCA scan
lines n-13 through n+13. If you have already read one scan line, then
reading a nearby scan line will be relatively fast because the library
can reuse some intermediate results from the YCA to RGBA conversion for
the first scan line.
RGBA to sub-sampled YCA conversion is a lot easier - and a bit faster -
if you can convert an entire frame at once. The conversion requires the
following steps:
- convert each RGBA pixel to YCA format
- low-pass filter and sub-sample the YCA pixels horizontally
- low-pass filter and sub-sample the YCA pixels vertically
The public header file ImfRgbaYca.h contains a set of helper functions
that perform these steps. The routines have been designed to minimize
generational loss when pixels are repeatedly converted back and forth
between RGBA and YCA.
Florian
Neall Verheyde wrote:
To reduce file size for our stereo-pair renders, I'm experimenting with YCA
encoded images.
Something like this is the goal:
channels (type chlist):
A, 16-bit floating-point, sampling 1 1
Y, 16-bit floating-point, sampling 1 1
RY, 16-bit floating-point, sampling 2 2
BY, 16-bit floating-point, sampling 2 2
right.A, 16-bit floating-point, sampling 1 1
right.Y, 16-bit floating-point, sampling 1 1
right.RY, 16-bit floating-point, sampling 2 2
right.BY, 16-bit floating-point, sampling 2 2
The problem I'm encountering is the RgbaOutputFile class does not seem to
support writing out more than four channels, and it does not support writing
out the multi-view prefix. The RgbaReader class seems to have some support for
the latter feature, although we are currently using the lower-level
Input/OutputFile classes for reading and writing. This suggests we have to
pack and unpack the YCA channels, with help from the functions in ImfRgbaYca.h
but doing this looks to be complicated if the private RgbaOutputFile::ToYca
helper class is any indication.
Is there a relatively painless way to create multiview YCA images without
having to cut-and-paste swaths of private implementation contained inside
ImfRgbaFile.cpp? I'm currently using OpenEXR 1.7.1.
Thanks,
Neall.
Gener8 Digital Media
_______________________________________________
Openexr-devel mailing list
Openexr-devel@nongnu.org
https://lists.nongnu.org/mailman/listinfo/openexr-devel
_______________________________________________
Openexr-devel mailing list
Openexr-devel@nongnu.org
https://lists.nongnu.org/mailman/listinfo/openexr-devel
