-----------------------------------------------------------------
May 7th 1998 FCD Version 0.0d
-----------------------------------------------------------------

INTEGRATORS:

Simon Winder
swinder@microsoft.com

-----------------------------------------------------------------
REASON FOR UPDATE:

Multiple changes to shape, texture, motion, and bug fixes.

-----------------------------------------------------------------
FILES CHANGED:


-----------------------------------------------------------------
COMMENTS:


This is the list of updates to the MS software and my comments:

A. Fixed numerous bugs for 12bit.
B. Fixed motion vector bug creating BVOP mismatch as noted by Jae-won Chung [jwchung98@hei.co.kr].
C. Updated code according to delta-FCD document sections.

1.1.25 padding for non 8-bit
Done. I updated the padding so that it would work ok with n-bit. However,
having spoken to Rockwell, I understand that there is no real intention to
allow the use of 12-bit at the same time as shape coding. In this case, it
is only the frame-based repeat padding that needs to work for n-bit. I have
updated both padding methods. Is there an intention to use shape and n-bit
together? What about greyscale alpha?

1.1.73 12 bit clipping limit
1.1.74 12-bit clipping limit
Done.

2.1.1 nextbits_bytealigned()
2.1.2 start code byte aligned
Done. No changes. I checked the current code and it looks ok. The decoder
always byte aligns before looking for a start code. The encoder does not
insert 8 bits between VO and VOL start codes.

2.1.3 add next_start_code() syntax
Done. There is no Visual Object syntax, so nothing is added here.

2.1.5 Displayable part of VOP
Done. No changes. The decoder already rounds up object layer width/height to
MB multiples.

2.1.6 vop_height and vop_width
Done. I don't think that a change is needed here.

2.1.7 random_access_vol
Done. No changes. This item was already handled correctly.

2.1.9 user_data()
Done. The findStartCode() method now skips any user data start code and
data.

1.2.1 remove acdc prediction condition
Done. All code for disabled acdc prediction is removed.

1.2.8 check rounding
Done. Rounding has been updated.

1.2.10 clipping limit
Done.

1.2.11 interlace and acdc prediction check
Done.

1.2.13 remove acdc flag
Done.

2.2.1 DC prediction at boundary blocks
Done. Unchanged, looks ok.

2.2.3 conditional (check sw)
2.2.4 conditional (check sw)
Done. I think there is no change. I could not understand the dfcd document
here.

2.2.5 ???
Done. Ignored.

2.2.12 default AC predictors
Done. Unchanged, looks ok.

2.2.13 just verify
Done.

2.2.14 12bit flc.
Done. I updated this so that it would work for the n-bit case.

1.4.7 padding of reference for 12-bit.
Done. But I understand that 12-bit is not meant to support shape.

2.3.1 mv predictors
Done. Unchanged, looks ok.

2.3.2 bab reference in bvop
Done. I did not allow the non coded vops to become reference vops though. I
think there is a problem here, which needs to be addressed by mpeg.

2.3.3 vector padding
Done.

2.3.9 default padding of non-coded reference vops
I did not do this because I think non-coded (transparent) vops should not be
reference vops (see 2.3.2 and my previous email).

2.3.10 cbpy tables
Done. This change is ridiculous. Neither Wei-ge nor myself can find any
start code emulation with the original (shorter) tables.

2.4.1 mv table entry
Done. Unchanged, looks ok.

Issues:

1. If the escape code FLCs for AC coefficients are expanded to allow for
12-bit DCT coeffs, then the tables for data partitioning mode should also be
expanded. I added both sets by extending the existing ones in the obvious
way, so that they would work for the general case of n-bit video.

2. Is 12-bit intended to support shape? I did not update this.

3. Is 12-bit intended to include chroma? I added some of the code. Rockwell
is going to test this.

4. Someone needs to look at the dumping of non-coded frames with temporal
scalability enabled, since I had trouble understanding the function of this
part of the software.

5. I did not change any code which would cause prediction to take place from
non-coded frames.

The major issue I think is #5, since if prediction can really be made from
non-coded frames, then this is a large software change.

Simon






