Skip to content

Commit c60a5a9

Browse files
minnendcopybara-github
minnend
authored and
copybara-github
committed
Adds more rate-distortion results for the minnen-2018-neurips model.
The model is from the paper: "Joint autoregressive and hierarchical priors for learned image compression" The new results include RD values for metrics that differ from the loss function (e.g. optimized for MS-SSIM and evaluated with PSNR). PiperOrigin-RevId: 405745486 Change-Id: I0cbb001edb40e446e03f5b2981b3dfd96554b8f1
1 parent 12b66d3 commit c60a5a9

16 files changed

+276
-0
lines changed

‎results/image_compression/kodak/MS-SSIM_sRGB_RGB/minnen-2018-neurips.txt renamed to ‎results/image_compression/kodak/MS-SSIM_sRGB_RGB/minnen-2018-neurips-opt-mse.txt

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77
# lambda value.
88
# 2. We often graph MS-SSIM values in dB for visual clarity using:
99
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
# 3. This model was optimized for RGB MSE.
1011
#
1112
# If you have questions or corrections, please contact:
1213
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).

‎results/image_compression/kodak/MS-SSIM_sRGB_RGB/minnen-2018-neurips-opt-msssim.txt

Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
# Aggregate rate-distortion data for "Minnen 2018 (NeurIPS)" 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+
# 3. This model was optimized for RGB MS-SSIM.
11+
#
12+
# If you have questions or corrections, please contact:
13+
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).
14+
15+
0.092495, 0.926154
16+
0.196784, 0.963019
17+
0.392549, 0.983832
18+
0.580614, 0.989581
19+
0.755953, 0.992419
20+
0.945877, 0.994319
21+
1.195109, 0.995809
22+
1.425766, 0.996772
23+
1.982975, 0.998191
24+
2.641110, 0.999101

‎results/image_compression/kodak/MS-SSIM_sRGB_Y/minnen-2018-neurips.txt renamed to ‎results/image_compression/kodak/MS-SSIM_sRGB_Y/minnen-2018-neurips-opt-mse.txt

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77
# lambda value.
88
# 2. We often graph MS-SSIM values in dB for visual clarity using:
99
# ms_ssim_db = -10 * log10(1 - ms_ssim).
10+
# 3. This model was optimized for RGB MSE.
1011
#
1112
# If you have questions or corrections, please contact:
1213
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).

‎results/image_compression/kodak/MS-SSIM_sRGB_Y/minnen-2018-neurips-opt-msssim.txt

Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
# Aggregate rate-distortion data for "Minnen 2018 (NeurIPS)" 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+
# 3. This model was optimized for RGB MS-SSIM.
11+
#
12+
# If you have questions or corrections, please contact:
13+
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).
14+
15+
0.092495, 0.934947
16+
0.196784, 0.968697
17+
0.392549, 0.987169
18+
0.580614, 0.992039
19+
0.755953, 0.994383
20+
0.945877, 0.995871
21+
1.195109, 0.997060
22+
1.425766, 0.997769
23+
1.982975, 0.998721
24+
2.641110, 0.999256

‎results/image_compression/kodak/PSNR_sRGB_RGB/minnen-2018-neurips.txt renamed to ‎results/image_compression/kodak/PSNR_sRGB_RGB/minnen-2018-neurips-opt-mse.txt

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,7 @@
55
# Notes:
66
# 1. Aggregate values were calculated by averaging over a constant
77
# lambda value.
8+
# 2. This model was optimized for RGB MSE.
89
#
910
# If you have questions or corrections, please contact:
1011
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).

‎results/image_compression/kodak/PSNR_sRGB_RGB/minnen-2018-neurips-opt-msssim.txt

Lines changed: 22 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,22 @@
1+
# Aggregate rate-distortion data for "Minnen 2018 (NeurIPS)" 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+
# 2. This model was optimized for RGB MS-SSIM.
9+
#
10+
# If you have questions or corrections, please contact:
11+
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).
12+
13+
0.092495, 25.376121
14+
0.196784, 27.127567
15+
0.392549, 29.273395
16+
0.580614, 30.408266
17+
0.755953, 31.429707
18+
0.945877, 32.262988
19+
1.195109, 33.434059
20+
1.425766, 34.352065
21+
1.982975, 35.366026
22+
2.641110, 36.833078

‎results/image_compression/kodak/PSNR_sRGB_Y/minnen-2018-neurips.txt renamed to ‎results/image_compression/kodak/PSNR_sRGB_Y/minnen-2018-neurips-opt-mse.txt

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,7 @@
55
# Notes:
66
# 1. Aggregate values were calculated by averaging over a constant
77
# lambda value.
8+
# 2. This model was optimized for RGB MSE.
89
#
910
# If you have questions or corrections, please contact:
1011
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).

‎results/image_compression/kodak/PSNR_sRGB_Y/minnen-2018-neurips-opt-msssim.txt

Lines changed: 22 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,22 @@
1+
# Aggregate rate-distortion data for "Minnen 2018 (NeurIPS)" on kodak.
2+
# The first column contains bits per pixel (bpp) values.
3+
# The second column contains PSNR/sRGB/Y' values.
4+
#
5+
# Notes:
6+
# 1. Aggregate values were calculated by averaging over a constant
7+
# lambda value.
8+
# 2. This model was optimized for RGB MS-SSIM.
9+
#
10+
# If you have questions or corrections, please contact:
11+
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).
12+
13+
0.092495, 25.649601
14+
0.196784, 27.390140
15+
0.392549, 29.542959
16+
0.580614, 30.716348
17+
0.755953, 31.758289
18+
0.945877, 32.604129
19+
1.195109, 33.786184
20+
1.425766, 34.708266
21+
1.982975, 35.722294
22+
2.641110, 37.149169

‎results/image_compression/tecnick/MS-SSIM_sRGB_RGB/minnen-2018-neurips-opt-mse.txt

Lines changed: 23 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,23 @@
1+
# Aggregate rate-distortion data for "Minnen 2018 (NeurIPS)" on tecnick.
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+
# 3. This model was optimized for RGB MSE.
11+
#
12+
# If you have questions or corrections, please contact:
13+
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).
14+
15+
0.066339, 0.929896
16+
0.126793, 0.956899
17+
0.198055, 0.969371
18+
0.300435, 0.978856
19+
0.433107, 0.984426
20+
0.615165, 0.987542
21+
0.891458, 0.992472
22+
1.507765, 0.993967
23+
2.425908, 0.998597

‎results/image_compression/tecnick/MS-SSIM_sRGB_RGB/minnen-2018-neurips-opt-msssim.txt

Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
# Aggregate rate-distortion data for "Minnen 2018 (NeurIPS)" on tecnick.
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+
# 3. This model was optimized for RGB MS-SSIM.
11+
#
12+
# If you have questions or corrections, please contact:
13+
# David Minnen (dminnen@google.com) or George Toderici (gtoderici@google.com).
14+
15+
0.080911, 0.951014
16+
0.160159, 0.974363
17+
0.306288, 0.986175
18+
0.454266, 0.990466
19+
0.603689, 0.991907
20+
0.746582, 0.993437
21+
0.933949, 0.996380
22+
1.138409, 0.997145
23+
1.642893, 0.998888
24+
2.288673, 0.999601

0 commit comments

Comments
 (0)