Skip to content

Conversation

@jwcullen
Copy link
Collaborator

  • Encoder bug-fixes improve this file's PSNR test performance.
  • Previously demixing matrices were applied incorrectly and led to bad scores in some channels.
  • Based on AOMediaCodec/iamf-tools@cdc930082

test_000070

When rendering from 7.1.4 to 7.1.2, the final two channels in reference file incorrectly were dropping channels instead of applying §3.8.2 gamma value.

Using the reference software, I ran this test before this PR:

code/test/tools/iamfdec/iamfdec -i0 -mp 42 -s10 -o3 tests/test_000070_rendered_id_42_sub_mix_0_layout_1_generated.wav -d 16 -r 48000 -disable_limiter tests/test_000070.iamf
python3 tests/psnr_calc.py --ref test_000070_rendered_id_42_sub_mix_0_layout_1.wav --target test_000070_rendered_id_42_sub_mix_0_layout_1_generated.wav --dir .

The original results had bad scores in the last two channels:

ch#0 PSNR: inf
[Abridged channels 1 through 7 (inf)]                                                                                                                                                                                       
ch#8 PSNR: 34.362047543353725 dB                                                                                                                                                                                       
ch#9 PSNR: 36.46787671489911 dB                                                                                                                                                                                        
average PSNR: 35.414962129126422      

And are significantly improved after:

ch#0 PSNR: inf
[Abridged channels 1 through 7 (inf)]                                                                                                                                                                                      
ch#8 PSNR: 99.48411673970611 dB
ch#9 PSNR: 99.36477363500742 dB
average PSNR: 99.424445187356767

test_000210

When rendering from 7.1.4 to 3.1.2, incorrect demixng matrices were applied in the reference file.

Results overall improve in several channels from:

ch#0 PSNR: 80.01661034738198 dB
ch#1 PSNR: 75.81802931912671 dB
ch#2 PSNR: inf
ch#3 PSNR: inf
ch#4 PSNR: 79.29201929325461 dB
ch#5 PSNR: 81.88812123962035 dB
average PSNR: 79.253695049845916

To:

ch#0 PSNR: 99.33922319776885 dB
ch#1 PSNR: 99.34121392154616 dB
ch#2 PSNR: inf
ch#3 PSNR: inf
ch#4 PSNR: 99.3462128493322 dB
ch#5 PSNR: 99.3433504401923 dB
average PSNR: 99.342500102209868

  - Encoder bug-fixes improve this file's PSNR test performance.
  - Previously demixing matrices were applied incorrectly and led to bad scores in some channels.
  - Based on AOMediaCodec/iamf-tools@cdc930082
@kavithaig
Copy link

Hello,
Will these two reference files be merged to the v1.0.1 repository also? I see the Source.zip for that tag has not been updated since last year Aug 26 2024 (https://github.com/AOMediaCodec/libiamf/releases/tag/v1.0.1). We are using v1.0.1 because we were told that was the latest for the base profile.

Basically I would like to make sure I pick up all the latest related to base profile based code update. I want to know how you will add new updates. Let me know if there is a mailing list I could be part of to know if there are any updates/fixes.

Thanks,
kavitha

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants