1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
|
13.7.2001
Friday 13th ... no comment -- Initial CVS release.
14.7.2001
Arghhh,... I should have wait a day. Everything was going wrong
and even went worse when I tried to fix it. In detail: I fixed
several Bugs in the Motion-Compensation. The most ugly bug was
*not* to use the reference frame but a subsampled one in the
full-pel search ... uhhhh ... That was bad ...
Several bugs in the block-copy-routines fixed (eg. that neat thing
overwriting a correct neighbor-block...)
Introduced a totally different method of deciding weather a block
is good or bad ... still need a better method here ! At least it
seems to work, now.
15.7.2001
Changed search to 8x8 blocks. (Not really. The search itself
remained (for the subsampled passes) at 16x16 and only 8x8 is used
for pel and half-pel search as well as for matching against the
SQD mean values. One odd thing is this seems to produces less
misdetected blocks the other is: *IF* we have visible blocks
they're of the same nature I have seen in MPEG4/DivX ...
Despite that it seems to work much better (and much slower).
Music keeps me alive ... Thanks to "Das Ich". and to "Jacobs Dröh-
nung"
|