VP8 Questions - Sigma Designs
Robert Pleva
Hello WebM team:
Our company is evaluating VP8 for possible support in our HW-based media processors, and our engineering team has identified some questions after reading the available specification documents. Could you please respond to the following questions? Thanks in advance….
1) Is there a set of conformance points (profile@level combinations) defined?
2) Are there conformance bitstreams associated with these conformance points?
The following questions are with reference to a 1080p30 high quality video stream:
3) What are the typical and highest bit-rates at the output (input) of an encoder (decoder)?
4) What worst case (i.e. highest) Boolean symbol input (output) rate does an encoder (decoder) have to sustain?
5) Section 7.2 describes a normalization process to keep range within the bounds 128 <= range <= 255: Is the number of “1-bit left-shifts” per second to normalize range almost equal to the bit-rate?
Regarding the alternate YUV color format mentioned in the VP8 spec:
“There are two different types of YUV color formats supported in the VP8 bitstream. They differ in terms of
their conversion from and to RGB color space. One corresponds to the traditional YUV color space similar to
the YCrCb color space defined in ITU-R BT.601. The other corresponds to a new YUV color space, whose
digital conversion from and to RGB can be implemented without multiplications and divides. VP8 Decoder
should decode and pass the information to the processes that convert the YUV output to RGB color space”
6) What specific information about the second YUV format mentioned in the text above would decoder need to pass to other processes?
Regards,
Director of Technical Marketing
Sigma Designs, Inc.
1778 McCarthy Blvd.
Milpitas, CA 95035
Phone: (408) 957-9767
Fax: (408) 957-9769
email: robert...@sdesigns.com
Paul Wilkins
Thanks for your questions. Some of these issues have been raised by
others and we are aware of the need to improve and extend the
specification in a more formal way.
1)-4) There are not currently any formal levels defined (in terms of
data rate per second / macro-blocks per second etc. and supported
image sizes). This is something that has been raised by several
hardware vendors. However there is a limit on the maximum image
dimensions (defined by the 14 bit height and width fields in the
header).
We have been discussing the idea of defining some levels and would
welcome your input. For example one vendor has already suggested that
they would prefer us to define limits on the number of bits / frame
for various different image sizes rather than just the number of bits
per second as defined in H.264 levels. Please feel free to add to this
debate. If we are going to do this I would rather we get things
defined sooner rather than later.
There are different encoder profiles that influence the decode
complexity. For example profile 1 indicates that sub-pixel filtering
should be done using a simpler bilinear filter and that the simple
loop filter should be used.
Yes we do have a set of conformance bit streams / test vectors. If we
decide to define restrictions in the form of levels etc. we will
insure that the test vector set is extended appropriately.
5) I guess the answer to this is yes... the number of << 1
normalisation cycles / second approximates to the bitrate (all be it
that there are a some non-bool coded bits).
6). I am not sure that you have the most up to date spec. The simpler
YUV formats is not supported in the baseline VP8 bitstream for WebM
(though there is still provision for it in the header).
Regards
Paul Wilkins
On Jun 8, 6:24 pm, "Robert Pleva" <Robert_Pl...@sdesigns.com> wrote:
> Hello WebM team:
>
> Our company is evaluating VP8 for possible support in our HW-based media
> processors, and our engineering team has identified some questions after
> reading the available specification documents. Could you please respond