Skip to content

Commit 8692f3c

Browse files
minnendcopybara-github
minnend
authored and
copybara-github
committed
Updates results for Balle (2017) and Balle (2018) for both MSE-optimized and MS-SSIM-optimized models (previously, all results were from the MSE-optimized model). This CL also adds PSNR results for Minnen (2020).
PiperOrigin-RevId: 312196405 Change-Id: Idc59c21fe1798c1fe3da7f03f20cfef12b6772d5
1 parent 832cd7a commit 8692f3c

38 files changed

+577
-31
lines changed

‎results/image_compression/README.md

Lines changed: 15 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -25,54 +25,58 @@ evaluation data sets are covered.
2525

2626
### Learning-based Methods
2727

28-
1. [Context-adaptive Entropy Model for End-to-end Optimized Image Compression](https://openreview.net/forum?id=HyxKIiAqYQ)\
28+
1. [Channel-wise autoregressive entropy models for learned image compression](http://research.minnen.org/papers/minnen-submitted-icip2020-draft.pdf)\
29+
David Minnen and Saurabh Singh\
30+
Int. Conf. on Image Processing (ICIP) 2020
31+
32+
2. [Context-adaptive Entropy Model for End-to-end Optimized Image Compression](https://openreview.net/forum?id=HyxKIiAqYQ)\
2933
Jooyoung Lee, Seunghyun Cho, and Seung-Kwon Beack\
3034
Int. Conf. on Learning Representations (ICLR) 2019
3135

32-
2. [Joint autoregressive and hierarchical priors for learned image
36+
3. [Joint autoregressive and hierarchical priors for learned image
3337
compression](https://arxiv.org/abs/1809.02736)\
3438
David Minnen, Johannes Ballé, and George Toderici\
3539
Advances in Neural Information Processing Systems (NeurIPS) 2018
3640

37-
3. [Learning a Code-Space Predictor by Exploiting Intra-Image-Dependencies](http://bmvc2018.org/contents/papers/0491.pdf)\
41+
4. [Learning a Code-Space Predictor by Exploiting Intra-Image-Dependencies](http://bmvc2018.org/contents/papers/0491.pdf)\
3842
Jan P. Klopp, Yu-Chiang Frank Wang, Shao-Yi Chien, and Liang-Gee Chen\
3943
British Machine Vision Conference (BMVC) 2018
4044

41-
4. [Variational Image Compression with a Scale Hyperprior](https://arxiv.org/abs/1802.01436)\
45+
5. [Variational Image Compression with a Scale Hyperprior](https://arxiv.org/abs/1802.01436)\
4246
Johannes Ballé, David Minnen, Saurabh Singh, Sung Jin Hwang, and Nick
4347
Johnston\
4448
Int. Conf. on Learning Representations (ICLR) 2018
4549

46-
5. [Image-dependent local entropy models for image compression with deep
50+
6. [Image-dependent local entropy models for image compression with deep
4751
networks](https://arxiv.org/abs/1805.12295)\
4852
David Minnen, George Toderici, Saurabh Singh, Sung Jin Hwang, and Michele
4953
Covell\
5054
Int. Conf. on Image Processing (ICIP) 2018
5155

52-
6. [Improved Lossy Image Compression With Priming and Spatially Adaptive Bit
56+
7. [Improved Lossy Image Compression With Priming and Spatially Adaptive Bit
5357
Rates for Recurrent Networks](https://arxiv.org/abs/1703.10114)\
5458
Nick Johnston, Damien Vincent, David Minnen, Michele Covell, Saurabh Singh,
5559
Troy Chinen, Sung Jin Hwang, Joel Shor, and George Toderici\
5660
IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) 2018
5761

58-
7. [Real-Time Adaptive Image Compression](https://arxiv.org/abs/1705.05823)\
62+
8. [Real-Time Adaptive Image Compression](https://arxiv.org/abs/1705.05823)\
5963
Oren Rippel and Lubomir Bourdev\
6064
International Conference on Machine Learning (ICML) 2017
6165

62-
8. [End-to-end Optimized Image Compression](https://arxiv.org/abs/1611.01704)\
66+
9. [End-to-end Optimized Image Compression](https://arxiv.org/abs/1611.01704)\
6367
Johannes Ballé, Valero Laparra, and Eero P. Simoncelli\
6468
Int. Conf. on Learning Representations (ICLR) 2017
6569

66-
9. [Lossy Image Compression with Compressive Autoencoders](https://openreview.net/forum?id=rJiNwv9gg)\
70+
10. [Lossy Image Compression with Compressive Autoencoders](https://openreview.net/forum?id=rJiNwv9gg)\
6771
Lucas Theis, Wenzhe Shi, Andrew Cunningham, and Ferenc Huszár\
6872
Int. Conf. on Learning Representations (ICLR) 2017
6973

70-
10. [Spatially adaptive image compression using a tiled deep network](https://arxiv.org/abs/1802.02629)\
74+
11. [Spatially adaptive image compression using a tiled deep network](https://arxiv.org/abs/1802.02629)\
7175
David Minnen, George Toderici, Michele Covell, Troy Chinen, Nick Johnston,
7276
Joel Shor, Sung Jin Hwang, Damien Vincent, and Saurabh Singh\
7377
Int. Conference on Image Processing (ICIP) 2017
7478

75-
11. [Full Resolution Image Compression with Recurrent Neural Networks](https://arxiv.org/abs/1608.05148)\
79+
12. [Full Resolution Image Compression with Recurrent Neural Networks](https://arxiv.org/abs/1608.05148)\
7680
George Toderici, Damien Vincent, Nick Johnston, Sung Jin Hwang, David
7781
Minnen, Joel Shor, and Michele Covell\
7882
IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017

‎results/image_compression/kodak/MS-SSIM_sRGB_RGB/balle-2017-iclr-opt-mse.txt

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
# Aggregate rate-distortion data for "Ballé 2017 (ICLR, opt. MSE)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains MS-SSIM/sRGB/R'G'B' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. We often graph MS-SSIM values in dB for visual clarity using:
9+
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
#
11+
# If you have questions or corrections, please contact:
12+
# David Minnen ([email protected]) or George Toderici ([email protected]).
13+
14+
0.119752, 0.903700
15+
0.194591, 0.931041
16+
0.316000, 0.954783
17+
0.481060, 0.969139
18+
0.721303, 0.980815
19+
1.060841, 0.986755
20+
1.458681, 0.992090
21+
1.957564, 0.994965

‎results/image_compression/kodak/MS-SSIM_sRGB_RGB/balle-2017-iclr-opt-msssim.txt

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
# Aggregate rate-distortion data for "Ballé 2017 (ICLR, opt. MS-SSIM)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains MS-SSIM/sRGB/R'G'B' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. We often graph MS-SSIM values in dB for visual clarity using:
9+
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
#
11+
# If you have questions or corrections, please contact:
12+
# David Minnen ([email protected]) or George Toderici ([email protected]).
13+
14+
0.088965, 0.911371
15+
0.159174, 0.944562
16+
0.274751, 0.966757
17+
0.456740, 0.981293
18+
0.657944, 0.988021
19+
1.039564, 0.993110
20+
1.519391, 0.996033
21+
2.100730, 0.997772

‎results/image_compression/kodak/MS-SSIM_sRGB_RGB/balle-2018-iclr-opt-mse.txt

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
# Aggregate rate-distortion data for "Ballé 2018 (ICLR, opt. MSE)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains MS-SSIM/sRGB/R'G'B' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. We often graph MS-SSIM values in dB for visual clarity using:
9+
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
#
11+
# If you have questions or corrections, please contact:
12+
# David Minnen ([email protected]) or George Toderici ([email protected]).
13+
14+
0.115239, 0.907527
15+
0.185698, 0.936307
16+
0.301804, 0.958691
17+
0.468972, 0.972416
18+
0.686378, 0.982478
19+
0.966864, 0.988344
20+
1.307441, 0.992647
21+
1.727503, 0.995267

‎results/image_compression/kodak/MS-SSIM_sRGB_RGB/balle-2018-iclr-opt-msssim.txt

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
# Aggregate rate-distortion data for "Ballé 2018 (ICLR, opt. MS-SSIM)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains MS-SSIM/sRGB/R'G'B' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. We often graph MS-SSIM values in dB for visual clarity using:
9+
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
#
11+
# If you have questions or corrections, please contact:
12+
# David Minnen ([email protected]) or George Toderici ([email protected]).
13+
14+
0.092031, 0.923012
15+
0.166818, 0.953873
16+
0.281376, 0.972845
17+
0.398321, 0.981048
18+
0.650651, 0.989613
19+
0.996335, 0.994012
20+
1.377538, 0.996171
21+
2.030330, 0.998020

‎results/image_compression/kodak/MS-SSIM_sRGB_Y/balle-2017-iclr-opt-mse.txt

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
# Aggregate rate-distortion data for "Ballé 2017 (ICLR, opt. MSE)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains MS-SSIM/sRGB/Y' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. We often graph MS-SSIM values in dB for visual clarity using:
9+
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
#
11+
# If you have questions or corrections, please contact:
12+
# David Minnen ([email protected]) or George Toderici ([email protected]).
13+
14+
0.119752, 0.912465
15+
0.194591, 0.938222
16+
0.316000, 0.960606
17+
0.481060, 0.973768
18+
0.721303, 0.984341
19+
1.060841, 0.989370
20+
1.458681, 0.994011
21+
1.957564, 0.996310

‎results/image_compression/kodak/MS-SSIM_sRGB_Y/balle-2017-iclr-opt-msssim.txt

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
# Aggregate rate-distortion data for "Ballé 2017 (ICLR, opt. MS-SSIM)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains MS-SSIM/sRGB/Y' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. We often graph MS-SSIM values in dB for visual clarity using:
9+
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
#
11+
# If you have questions or corrections, please contact:
12+
# David Minnen ([email protected]) or George Toderici ([email protected]).
13+
14+
0.088965, 0.921848
15+
0.159174, 0.952119
16+
0.274751, 0.972189
17+
0.456740, 0.984918
18+
0.657944, 0.990717
19+
1.039564, 0.994777
20+
1.519391, 0.997142
21+
2.100730, 0.998423

‎results/image_compression/kodak/MS-SSIM_sRGB_Y/balle-2018-iclr-opt-mse.txt

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
# Aggregate rate-distortion data for "Ballé 2018 (ICLR, opt. MSE)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains MS-SSIM/sRGB/Y' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. We often graph MS-SSIM values in dB for visual clarity using:
9+
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
#
11+
# If you have questions or corrections, please contact:
12+
# David Minnen ([email protected]) or George Toderici ([email protected]).
13+
14+
0.115239, 0.916089
15+
0.185698, 0.943538
16+
0.301804, 0.964504
17+
0.468972, 0.976787
18+
0.686378, 0.985790
19+
0.966864, 0.991048
20+
1.307441, 0.994607
21+
1.727503, 0.996637

‎results/image_compression/kodak/MS-SSIM_sRGB_Y/balle-2018-iclr-opt-msssim.txt

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
# Aggregate rate-distortion data for "Ballé 2018 (ICLR, opt. MS-SSIM)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains MS-SSIM/sRGB/Y' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. We often graph MS-SSIM values in dB for visual clarity using:
9+
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
#
11+
# If you have questions or corrections, please contact:
12+
# David Minnen ([email protected]) or George Toderici ([email protected]).
13+
14+
0.092031, 0.932611
15+
0.166818, 0.960943
16+
0.281376, 0.977778
17+
0.398321, 0.984935
18+
0.650651, 0.992063
19+
0.996335, 0.995652
20+
1.377538, 0.997328
21+
2.030330, 0.998616

‎results/image_compression/kodak/PSNR_sRGB_RGB/balle-2017-iclr-opt-mse.txt

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,19 @@
1+
# Aggregate rate-distortion data for "Ballé 2017 (ICLR, opt. MSE)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains PSNR/sRGB/R'G'B' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
#
9+
# If you have questions or corrections, please contact:
10+
# David Minnen ([email protected]) or George Toderici ([email protected]).
11+
12+
0.119752, 26.775134
13+
0.194591, 28.348719
14+
0.316000, 30.020793
15+
0.481060, 31.729556
16+
0.721303, 33.685797
17+
1.060841, 35.815864
18+
1.458681, 38.019954
19+
1.957564, 40.133996

0 commit comments

Comments
 (0)