Subversion Repositories mpeg2fpga

The following changes have been made to debug spatial scalability:

gethdr.c

--------

Temporal_reference is used to compute the frame number of each frame,

named true_framenum.  The periodic reset at each GOP header as well as

the wrap of temporal_reference at 1024 cause a base value

temp_ref_base to be incremented accordingly.

spatscal.c

----------

getspatref()

A potential problem: Variable char fname[32] was dimensioned

statically and too small.

true_framenum is used instead of lower_layer_temporal_reference to

determine the lower layer frame to be read for spatial prediction.

The verification of lower_layer_temporal_reference is not possible

since the temporal reference values that have been encoded into the

base layer bitstream are not available to the enhancement layer

decoder.

Since there is no decoder timing information available, the rules on

which frames can legally be used as spatial prediction frames cannot

be checked.

Lower layer frames are read field-wise or frame-wise, depending on the

lower_layer_progressive_frame flag. Consistency between layers is

checked since the file format for frame and field pictures differs.

Note that the base layer decoder must not use the -f option to enforce

frame-wise storage.

Note further that only yuv image format (option -o0) is supported as

input format.

spatpred()

The code for the various combinations of llprog_frame, llfieldsel and

prog_frame has been completed and verified with the tceh_conf23

bitstream that uses all permissive combinations.

getpic.c

--------

A small bug when storing an I- or P-frame: The prog_frame flag that

the decoder knows when storing the oldrefframe belongs to the current

refframe. Therefore the old value of the flag needs to be memorized.

store.c

-------

A potential problem: the filename variables char outname[32],

tmpname[32] are statically dimensioned and quite small.

The concept of time in this video decoder software

--------------------------------------------------

When decoding a non-scalable bitstream, the frame number (i.e.

temporal position) of the current I- or P-frame can be derived

implicitly from the number of preceding B-frames after they have been

decoded. Therefore the temporal_reference entry in the picture header

is somewhat redundant and does not necessarily have to be evaluated in

the decoding process.

Decoding of the enhancement layer of a spatial scalable hierarchy,

however, requires to know the temporal position of each frame at the

instant when it is decoded, since data from a lower layer reference

frame has to be incorporated.

In the architecture of this video-only decoder decoding of a spatial

scalable hierarchy of bitstreams is done by calling mpeg2decode once

for the base layer bitstream and a second time for the enhancement

layer bitstream, indicating where the decoded base layer frames can be

found (option -s).

Here the concept of time is only present in the form of frame numbers.

Therefore spatial scalable bitstream hierarchies can only be handled

under the assumption that base and enhancement layer bitstreams are

decoded to image sequences where corresponding images of both layers

have identical frame numbers.

More specifically this means that base and enhancement layer

bitstreams must contain video with the same frame rate. Furthermore

only the temporally coincident frame of the base layer can be accessed

for spatial prediction by the enhancement layer decoder, since it is

not possible to resolve unambiguously the lower_layer_temporal_reference

which is meant to further specify the lower layer reference frame.

======================== SPATIAL.DOC ========================0

Decoding a spatial scalable hierarchy of bitstreams

---------------------------------------------------

With this video-only decoder decoding of a spatial scalable hierarchy

of bitstreams is done by calling mpeg2decode once for the base layer

bitstream and a second time for the enhancement layer bitstream,

indicating where the decoded base layer frames can be found

(using option -s and supplying ).

mpeg2decode -r -o0 base.mpg base%d%c

mpeg2decode -r -o0 -f -s base%d%c enh.mpg enh%d

Note that the base layer decoder must not use the -f option to enforce

frame-wise storage.

Note further that only yuv image format (option -o0) is supported as

input format.

Timing / layer synchronisation in this video decoder software

-------------------------------------------------------------

When decoding a non-scalable bitstream, the frame number (i.e.

temporal position) of the current I- or P-frame can be derived

implicitly from the number of preceding B-frames after they have been

decoded. Therefore the temporal_reference entry in the picture header

is somewhat redundant and does not necessarily have to be evaluated in

the decoding process.

Decoding of the enhancement layer of a spatial scalable hierarchy,

however, requires to know the temporal position of each frame at the

instant when it is decoded, since data from a lower layer reference

frame has to be incorporated.

The concept of time is only present in the form of frame numbers.

Therefore spatial scalable bitstream hierarchies can only be handled

under the assumption that base and enhancement layer bitstreams are

decoded to image sequences where corresponding images of both layers

have identical frame numbers.

More specifically this means that base and enhancement layer

bitstreams must contain video with the same frame rate. Furthermore

only the temporally coincident frame of the base layer can be accessed

for spatial prediction by the enhancement layer decoder, since it is

not possible to resolve unambiguously the lower_layer_temporal_reference

which is meant to further specify the lower layer reference frame.

Lower layer frames are read field-wise or frame-wise, depending on the

lower_layer_progressive_frame flag. Consistency between layers in this

respect is checked since the file format for frame and field pictures

differs.