avcodec/jpeg2000: Fix FF_DWT97_INT to pass the conformance testing defined in ISO/IEC 15444-4 #12
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
osamu620
force-pushed
the
patches/fix-FF_DWT97_INT-v6
branch
from
bdc2d08
to
fcd1485
Compare
osamu620
commented
Oct 21, 2024
| @@ -268,7 +268,7 @@ static void init_band_stepsize(AVCodecContext *avctx, | |||
| av_log(avctx, AV_LOG_ERROR, "stepsize out of range\n"); | |||
| } | |||
|
|
|||
| band->i_stepsize = lrint(band->f_stepsize * (1 << 15) + 0.5f); | |||
| band->i_stepsize = (int)floorf(band->f_stepsize * (1 << 15)); | |||
There was a problem hiding this comment.
@palemieux With this correction, I got the following results.
❯ ./dwt97-diff.sh ~/Documents/data/images/ppm/ElephantDream_small.ppm 0
.rw-rw-r-- 1,497,613 osamu 21 Oct 16:39 encoded-HEAD.j2c
.rw-rw-r-- 1,498,160 osamu 21 Oct 16:39 encoded-PATCH.j2c
PSNR decoded by exact-HEAD
90.2224 (0.902224)
PSNR decoded by exact-PATCH
91.6198 (0.916198)
PSNR decoded by float-HEAD
79.0372 (0.790372)
PSNR decoded by float-PATCH
81.9933 (0.819933)
Now, the mismatch between float97 and int97 seems to have disappeared. For both float and int, the larger codestream size gives better quality.
There was a problem hiding this comment.
Looks great. Two questions:
- why move
#define F_LFTG_ALPHAto a difference source file? - why use
//sometimes and/*some other times?
There was a problem hiding this comment.
Thanks for reviewing!
-
A1: I thought
jpeg2000dwt.hwas included intests/jpeg2000dwt.c. But I found thatjpeg2000dwt.cwas included. So, this change makes no sense. I would revert this in the next commit. -
A2: No reason exists. In the next commit, I will unify the commenting style into
//.
osamu620
force-pushed
the
patches/fix-FF_DWT97_INT-v6
branch
3 times, most recently
from
0eed928
to
1a4e3be
Compare
…fined in ISO/IEC 15444-4 Fix for the integer version of the inverse 9-7 DWT processing (FF_DWT97_INT, https://trac.ffmpeg.org/ticket/10123), which is activated with `-flags +bitexact`. I went through the code path for the DWT 9-7 transform (integer) and improved precision to match conformance codestream. As a result, the encoded codestream size is slightly larger for a given Q value. For example, `-flags +bitexact -i lena.pnm -q: 20 -format j2k -y tmp.j2c` gives 13K (HEAD) and 19K (with this patch). This commit also updates the source and reference files for affected FATE tests.
osamu620
force-pushed
the
patches/fix-FF_DWT97_INT-v6
branch
from
1a4e3be
to
3227b41
Compare
palemieux
requested changes
Oct 24, 2024
| @@ -268,7 +268,7 @@ static void init_band_stepsize(AVCodecContext *avctx, | |||
| av_log(avctx, AV_LOG_ERROR, "stepsize out of range\n"); | |||
| } | |||
|
|
|||
| band->i_stepsize = lrint(band->f_stepsize * (1 << 15) + 0.5f); | |||
| band->i_stepsize = (int)floorf(band->f_stepsize * (1 << 15)); | |||
There was a problem hiding this comment.
Looks great. Two questions:
- why move
#define F_LFTG_ALPHAto a difference source file? - why use
//sometimes and/*some other times?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
avcodec/jpeg2000: Fix FF_DWT97_INT to pass the conformance testing defined in ISO/IEC 15444-4
Fix for the integer version of the inverse 9-7 DWT processing (FF_DWT97_INT, https://trac.ffmpeg.org/ticket/10123), which is activated with
-flags +bitexact.I went through the code path for the DWT 9-7 transform (integer) and improved precision to match conformance codestream.
As a result, the encoded codestream size is slightly larger for a given Q value. For example,
-flags +bitexact -i lena.pnm -q: 20 -format j2k -y tmp.j2cgives 13K (HEAD) and 19K (with this patch).This commit also updates the source and reference files for affected FATE tests.