Performance

Below are performance measurements of each augmenter for image augmentation (augment_images()), heatmap augmentation (augment_heatmaps()) and keypoint/landmark augmentation (augment_keypoints()). (Last updated for 0.3.0)

System: The numbers were computed based on a haswell-generation i7 3.2Ghz CPU with DDR3 memory. That is a rather dated system by today’s standards. A modern, high-end system should achieve higher bandwidths.

All experiments were conducted using python 3.7 and numpy 1.17.0. Note that the precise python/numpy version can have significant impact on your performance.

Experiments Settings: All augmenters were run with reasonable parameter choices that should reflect expected real-world usage, while avoiding too simple parameter values that would lead to inflated scores. Some parameter choices are listed below, the remaining ones can be looked up in measure_performance.py. Kernel sizes were all set to 3x3, unless otherwise mentioned. The inputs focused on a small and large image-size setting, using 64x64x3 and 224x224x3 as the respective sizes. The base image was taken from skimage.data.astronaut, which should be a representative real-world image. Batch sizes of 1 and 128 were tested. Each augmenter was run at least 40 times on the generated input and the average of the measured runtimes was computed to derive bandwidth in mbit per second and th raw number of augmented items (e.g. images) per second.

Results Overview

From the results, the following points can be derived.

Inputs:

  • Use large batch sizes whenever possible. Many augmenters are significantly faster with these.
  • Large image sizes lead to higher throughput based on mbit/sec. Smaller images lead to lower throughput, but significantly more items/sec (roughly 4-10x more). Use small images whenever possible.
  • For keypoint-based and heatmap-based augmentation, try to increase the number of items per augmented instance. E.g. augment_keypoints() accepts a list of KeypointsOnImage instances, with each such instance representing the keypoints on an image. Try to place for each image all keypoints in the respective KeypointsOnImage instance instead of splitting them into multiple such instances (which would be more work anyways). The same is true for bounding boxes, heatmaps and segmentation maps.
  • Keypoint- and heatmap-based inputs are only affected by augmenters that change the geometry of the image (e.g. Crop or Affine). Other augmenters are essentially free to execute as they do not perform any changes.
  • Keypoint-based augmentation is very fast for almost all augmenters, reaching several 100k keypoints per second. Slower augmenters are ElasticTransformation and PiecewiseAffine, as these currently have to fall back to image-based algorithms.

Parameter choices:

  • When possible, nearest neighbour interpolation or linear interpolation should be used as these are significantly faster than other options. Most augmenters that use interpolation offer either an order parameter (0=nearest neighbour, 1=linear) or an interpolation parameter (“nearest”, “linear”).
  • Using keep_size=True is the default setting in all augmenters that change image sizes. It is convenient, as it ensures that image sizes are not altered by the augmentation. It does however incur a significant performance penalty, often more than halving the bandwidth. Try keep_size=False when possible. You can still resize images manually after augmentation or by using KeepSizeByResize(Sequential(<augmenters>)).
  • When augmenters offer modes to fill newly created pixels in user-defined ways (e.g. pad_mode=constant in Pad to fill up all padded pixels with a specified constant color), using edge instead of constant will usually not incur a significant performance penalty.

Specific Augmenter suggestions:

  • For augmenters where an elementwise sibling exists (e.g. Multiply and MultiplyElementwise), the elementwise augmenter is usually significantly slower than the non-elementwise one.
  • If blurring is required, AverageBlur is the fastest choice, followed by GaussianBlur.
  • Augmenters that operate on coarser images (e.g. CoarseDropout vs Dropout) can be significantly faster than their non-coarse siblings.
  • Contrast normalizing augmenters are all comparable in performance, except for histogram-based ones, which are significantly slower.
  • PiecewiseAffine is a very slow augmenter and should usually be replaced by ElasticTransformation, which achieves similar outputs and is quite a bit faster.
  • Superpixels is a fairly slow augmenter and should usually be wrapped in e.g. Sometimes to not apply it very often and reduce its performance impact.
  • Weather augmenters other than FastSnowyLandscape are rather slow and should only be used when sensible.

Images

Numbers below are for small images (64x64x3) and large images (224x224x3). B=1 denotes a batch size of 1, B=128 one of 128.

In mbit/sec:

  64x64x3, uint8 224x224x3, uint8
Augmenter B=1 B=128 B=1 B=128
Sequential (2xNoop) 10291.8 51537.8 37374.1 46140.5
Sequential (2xNoop, random_order) 1160.2 37697.3 11243.0 45456.6
SomeOf (1-3, 3xNoop) 286.2 1588.5 2671.4 4851.3
SomeOf (1-3, 3xNoop, random_order) 239.2 1633.3 2277.9 3984.6
OneOf (3xNoop) 786.0 1994.0 6091.6 8357.4
Sometimes (Noop) 671.1 15784.4 6165.2 16099.6
WithChannels ([1,2], Noop) 2103.2 10006.9 7839.0 8354.2
Noop 16708.9 53947.0 48487.4 48685.3
Lambda (return input) 16040.9 54141.6 45396.7 51521.4
AssertLambda (return True) 15666.0 53733.9 46598.6 48569.5
AssertShape (None, H, W, None) 6315.4 18876.3 29643.0 44398.4
ChannelShuffle (0.5) 528.8 1605.4 3653.2 5335.8
Add 276.1 573.3 2560.3 4034.8
AddElementwise 265.8 319.3 990.9 1006.1
AdditiveGaussianNoise 206.1 237.7 749.9 755.3
AdditiveLaplaceNoise 171.6 193.0 441.6 438.8
AdditivePoissonNoise 145.4 156.7 310.4 307.6
Multiply 360.8 1014.7 3074.5 5610.4
MultiplyElementwise 239.4 288.1 983.5 1002.7
Dropout (1-5%) 303.3 379.5 1121.2 1175.2
CoarseDropout (1-5%, size=1-10%) 165.8 185.7 1171.7 1203.8
ReplaceElementwise 152.8 169.9 738.8 752.8
ImpulseNoise 130.1 141.7 494.1 494.6
SaltAndPepper 136.7 150.2 695.5 701.0
CoarseSaltAndPepper 99.6 105.7 693.6 713.2
Salt 113.5 121.0 622.4 629.5
CoarseSalt 86.7 90.7 640.0 642.1
Pepper 112.6 120.9 620.9 628.9
CoarsePepper 86.5 91.0 642.5 640.8
Invert (10%) 664.6 12262.6 6715.9 30743.9
JpegCompression (50-99%) 80.1 91.9 300.0 309.4
Alpha (Noop) 264.2 506.4 918.8 939.8
AlphaElementwise (Noop) 188.9 215.8 471.1 442.5
SimplexNoiseAlpha (Noop) 29.8 28.3 187.9 175.1
FrequencyNoiseAlpha (Noop) 37.1 36.0 216.7 208.2
GaussianBlur (sigma=(1,5)) 283.4 629.4 2367.5 3663.1
AverageBlur 435.4 3457.5 3101.8 6276.9
MedianBlur 173.2 265.2 306.1 313.9
BilateralBlur 158.2 366.5 447.9 489.6
MotionBlur 74.7 75.2 703.1 719.4
WithColorspace (HSV, Noop) 695.5 1067.5 1444.4 1527.1
WithHueAndSaturation 229.7 335.8 646.7 664.1
MultiplyHueAndSaturation 99.3 150.1 410.9 446.6
MultiplyHue 93.4 151.9 431.8 477.0
MultiplySaturation 93.8 152.5 427.2 479.1
AddToHueAndSaturation 228.3 768.0 944.5 1210.5
AddToHue 267.3 769.4 1002.7 1211.8
AddToSaturation 269.1 767.3 985.8 1208.6
ChangeColorspace (HSV) 440.1 800.9 1641.4 1868.2
Grayscale 208.8 332.3 680.6 728.5
KMeansColorQuantization (2-16 colors) 23.9 40.9 209.3 211.7
UniformColorQuantization (2-16 colors) 187.3 327.1 638.7 735.2
GammaContrast 259.6 325.1 2319.9 2754.9
SigmoidContrast 206.5 245.9 1974.1 2237.4
LogContrast 257.0 325.8 2334.9 2764.3
LinearContrast 324.3 430.9 2743.4 3360.2
AllChannelsHistogramEqualization 1110.6 1912.8 2191.6 2302.4
HistogramEqualization 470.5 878.1 1187.3 1218.7
AllChannelsCLAHE 143.2 344.8 995.4 1515.4
CLAHE 136.9 432.1 766.9 1035.4
Convolve (3x3) 1303.6 2820.4 4369.1 4833.5
Sharpen 261.0 295.0 1708.7 1828.9
Emboss 261.9 296.1 1769.5 1904.3
EdgeDetect 383.4 459.2 2260.1 2452.5
DirectedEdgeDetect 99.6 102.0 890.8 913.8
Canny 63.0 109.7 295.5 338.1
Fliplr (p=100%) 1165.1 4625.7 5828.7 7417.2
Flipud (p=100%) 1468.1 13368.8 8842.6 14318.9
Affine (order=0, constant) 96.3 272.4 943.3 1957.6
Affine (order=1, constant) 93.5 247.3 840.3 1555.0
Affine (order=3, constant) 84.6 206.1 498.6 682.7
Affine (order=1, edge) 93.3 246.3 830.6 1512.6
Affine (order=1, constant, skimage) 49.9 74.5 178.7 198.6
PiecewiseAffine (4x4, order=1, constant) 5.0 4.9 27.4 27.3
PiecewiseAffine (4x4, order=0, constant) 5.2 5.1 33.2 33.6
PiecewiseAffine (4x4, order=1, edge) 5.0 4.9 27.5 28.3
PiecewiseAffine (8x8, order=1, constant) 1.1 1.1 9.2 9.5
PerspectiveTransform 155.9 221.3 1129.9 1363.8
PerspectiveTransform (keep_size) 134.2 178.5 831.7 968.8
ElasticTransformation (order=0, constant) 108.2 183.0 572.1 726.5
ElasticTransformation (order=1, constant) 102.8 168.6 531.7 656.2
ElasticTransformation (order=1, nearest) 104.1 169.6 532.7 655.4
ElasticTransformation (order=1, reflect) 102.5 168.6 526.5 654.5
Rot90 463.3 4300.0 4820.7 25573.1
Rot90 (keep_size) 400.2 2267.4 2398.0 3479.9
AveragePooling 134.5 187.4 481.4 519.3
AveragePooling (keep_size) 119.7 157.3 422.2 472.5
MaxPooling 143.7 202.8 518.6 584.1
MaxPooling (keep_size) 127.3 168.3 476.1 525.2
MinPooling 144.3 197.1 518.0 572.6
MinPooling (keep_size) 128.5 165.6 496.8 521.4
MedianPooling 133.5 178.5 607.7 698.3
MedianPooling (keep_size) 118.7 151.8 573.2 617.8
Superpixels (max_size=64, cubic) 10.4 10.8 124.7 123.6
Superpixels (max_size=64, linear) 11.0 10.8 131.4 124.1
Superpixels (max_size=128, linear) 10.8 10.9 52.7 53.0
Superpixels (max_size=224, linear) 10.5 11.1 20.0 20.4
UniformVoronoi (250-1000k points, linear) 3.4 3.4 10.4 10.6
RegularGridVoronoi (16-31 rows/cols) 3.4 3.4 10.7 10.8
RelativeRegularGridVoronoi (7%-14% rows/cols) 3.5 3.5 3.5 3.6
Resize (nearest) 302.7 780.1 2436.5 4445.9
Resize (linear) 287.4 679.6 1802.3 2779.4
Resize (cubic) 267.9 586.8 1357.7 1848.8
CropAndPad 208.9 228.8 2099.8 2335.9
CropAndPad (edge) 209.0 231.2 2105.4 2312.0
CropAndPad (keep_size) 169.2 178.5 1278.1 1291.0
Crop 331.1 381.7 3198.3 3683.7
Crop (keep_size) 242.0 261.9 1700.4 1724.0
Pad 208.1 226.8 2009.4 2238.7
Pad (edge) 207.3 227.4 1977.3 2184.8
Pad (keep_size) 167.7 176.3 1197.4 1205.8
PadToFixedSize 202.1 1330.1 2134.7 7743.2
CropToFixedSize 388.5 3856.7 3747.3 11881.7
KeepSizeByResize (CropToFixedSize(nearest)) 203.7 1032.9 1746.1 3778.9
KeepSizeByResize (CropToFixedSize(linear)) 196.7 884.4 1406.0 2555.9
KeepSizeByResize (CropToFixedSize(cubic)) 186.8 745.9 1120.7 1741.7
FastSnowyLandscape 157.5 270.3 514.1 574.3
Clouds 19.9 20.1 60.8 60.1
Fog 33.9 33.9 99.8 99.4
CloudLayer 33.1 33.2 99.0 99.1
Snowflakes 16.4 16.7 87.9 94.9
SnowflakesLayer 33.1 33.6 192.0 191.2

In images/sec:

  64x64x3, uint8 224x224x3, uint8
Augmenter B=1 B=128 B=1 B=128
Sequential (2xNoop) 109779.4 549736.6 32543.4 40176.8
Sequential (2xNoop, random_order) 12375.0 402104.3 9789.9 39581.3
SomeOf (1-3, 3xNoop) 3053.2 16944.4 2326.1 4224.3
SomeOf (1-3, 3xNoop, random_order) 2551.2 17421.4 1983.5 3469.6
OneOf (3xNoop) 8384.3 21269.4 5304.3 7277.2
Sometimes (Noop) 7158.0 168366.5 5368.3 14018.7
WithChannels ([1,2], Noop) 22434.6 106739.9 6825.8 7274.4
Noop 178228.2 575434.6 42220.3 42392.6
Lambda (return input) 171103.0 577510.3 39529.1 44862.2
AssertLambda (return True) 167103.7 573161.6 40575.6 42291.8
AssertShape (None, H, W, None) 67363.9 201347.2 25811.6 38659.8
ChannelShuffle (0.5) 5640.1 17123.9 3181.1 4646.2
Add 2945.4 6114.7 2229.4 3513.3
AddElementwise 2835.7 3406.0 862.9 876.1
AdditiveGaussianNoise 2197.9 2535.8 653.0 657.7
AdditiveLaplaceNoise 1830.4 2058.9 384.6 382.1
AdditivePoissonNoise 1551.4 1671.6 270.3 267.8
Multiply 3848.9 10823.4 2677.1 4885.2
MultiplyElementwise 2553.6 3072.9 856.4 873.1
Dropout (1-5%) 3235.3 4047.5 976.2 1023.3
CoarseDropout (1-5%, size=1-10%) 1768.0 1980.8 1020.3 1048.2
ReplaceElementwise 1630.1 1812.7 643.3 655.5
ImpulseNoise 1387.5 1511.0 430.2 430.7
SaltAndPepper 1458.0 1602.4 605.6 610.4
CoarseSaltAndPepper 1062.3 1128.0 604.0 621.1
Salt 1210.5 1290.3 542.0 548.1
CoarseSalt 925.1 967.4 557.3 559.1
Pepper 1201.0 1289.8 540.6 547.6
CoarsePepper 922.2 970.6 559.4 558.0
Invert (10%) 7089.3 130801.3 5847.8 26770.2
JpegCompression (50-99%) 854.3 980.4 261.2 269.4
Alpha (Noop) 2818.0 5401.1 800.1 818.4
AlphaElementwise (Noop) 2015.3 2301.7 410.2 385.3
SimplexNoiseAlpha (Noop) 317.8 301.8 163.6 152.4
FrequencyNoiseAlpha (Noop) 395.5 384.3 188.7 181.3
GaussianBlur (sigma=(1,5)) 3023.1 6713.1 2061.5 3189.6
AverageBlur 4643.9 36880.1 2700.9 5465.6
MedianBlur 1847.1 2829.0 266.5 273.4
BilateralBlur 1687.5 3909.4 390.0 426.3
MotionBlur 797.3 801.9 612.2 626.4
WithColorspace (HSV, Noop) 7418.5 11386.4 1257.7 1329.8
WithHueAndSaturation 2450.6 3581.9 563.1 578.3
MultiplyHueAndSaturation 1058.9 1601.1 357.8 388.8
MultiplyHue 996.2 1620.1 376.0 415.3
MultiplySaturation 1000.5 1626.7 372.0 417.2
AddToHueAndSaturation 2435.2 8192.2 822.4 1054.1
AddToHue 2851.1 8207.3 873.1 1055.2
AddToSaturation 2870.5 8184.4 858.4 1052.4
ChangeColorspace (HSV) 4694.4 8542.6 1429.2 1626.8
Grayscale 2227.6 3544.3 592.6 634.4
KMeansColorQuantization (2-16 colors) 255.2 436.1 182.2 184.3
UniformColorQuantization (2-16 colors) 1997.6 3489.2 556.1 640.2
GammaContrast 2769.1 3467.9 2020.0 2398.8
SigmoidContrast 2202.9 2623.4 1719.0 1948.2
LogContrast 2740.9 3474.9 2033.1 2407.0
LinearContrast 3459.0 4596.5 2388.8 2925.9
AllChannelsHistogramEqualization 11846.3 20403.2 1908.3 2004.8
HistogramEqualization 5019.1 9366.5 1033.8 1061.2
AllChannelsCLAHE 1527.9 3678.2 866.7 1319.5
CLAHE 1459.9 4609.3 667.8 901.6
Convolve (3x3) 13905.2 30084.2 3804.4 4208.8
Sharpen 2784.0 3146.7 1487.8 1592.5
Emboss 2793.5 3158.7 1540.8 1658.2
EdgeDetect 4089.5 4897.9 1968.0 2135.5
DirectedEdgeDetect 1062.8 1088.3 775.7 795.7
Canny 671.8 1169.9 257.3 294.4
Fliplr (p=100%) 12427.9 49341.2 5075.3 6458.5
Flipud (p=100%) 15659.5 142600.4 7699.7 12468.1
Affine (order=0, constant) 1026.9 2906.0 821.4 1704.6
Affine (order=1, constant) 997.7 2638.0 731.7 1354.0
Affine (order=3, constant) 902.0 2198.5 434.2 594.5
Affine (order=1, edge) 995.6 2626.9 723.2 1317.1
Affine (order=1, constant, skimage) 532.0 794.7 155.6 172.9
PiecewiseAffine (4x4, order=1, constant) 53.5 52.1 23.9 23.7
PiecewiseAffine (4x4, order=0, constant) 55.3 54.2 28.9 29.3
PiecewiseAffine (4x4, order=1, edge) 53.4 52.4 23.9 24.7
PiecewiseAffine (8x8, order=1, constant) 12.1 11.8 8.0 8.3
PerspectiveTransform 1663.0 2360.1 983.9 1187.5
PerspectiveTransform (keep_size) 1431.2 1904.2 724.2 843.6
ElasticTransformation (order=0, constant) 1154.1 1952.2 498.2 632.6
ElasticTransformation (order=1, constant) 1096.4 1798.2 463.0 571.3
ElasticTransformation (order=1, nearest) 1110.1 1809.5 463.8 570.7
ElasticTransformation (order=1, reflect) 1093.3 1798.3 458.4 569.9
Rot90 4942.1 45866.6 4197.6 22267.7
Rot90 (keep_size) 4268.9 24186.1 2088.0 3030.2
AveragePooling 1434.7 1999.3 419.2 452.2
AveragePooling (keep_size) 1276.9 1678.1 367.6 411.4
MaxPooling 1533.3 2162.8 451.6 508.6
MaxPooling (keep_size) 1358.2 1795.6 414.6 457.3
MinPooling 1539.0 2102.2 451.1 498.6
MinPooling (keep_size) 1370.6 1766.1 432.6 454.0
MedianPooling 1424.2 1903.7 529.2 608.0
MedianPooling (keep_size) 1266.0 1619.0 499.1 537.9
Superpixels (max_size=64, cubic) 111.3 115.7 108.6 107.6
Superpixels (max_size=64, linear) 117.4 115.2 114.4 108.1
Superpixels (max_size=128, linear) 115.6 116.3 45.9 46.2
Superpixels (max_size=224, linear) 112.0 118.2 17.4 17.8
UniformVoronoi (250-1000k points, linear) 36.4 36.2 9.1 9.2
RegularGridVoronoi (16-31 rows/cols) 36.6 36.2 9.3 9.4
RelativeRegularGridVoronoi (7%-14% rows/cols) 37.7 37.2 3.1 3.1
Resize (nearest) 3229.3 8321.3 2121.6 3871.3
Resize (linear) 3065.2 7248.8 1569.4 2420.2
Resize (cubic) 2857.5 6259.3 1182.2 1609.8
CropAndPad 2228.8 2440.1 1828.4 2034.0
CropAndPad (edge) 2229.1 2465.8 1833.3 2013.1
CropAndPad (keep_size) 1804.6 1903.5 1112.9 1124.2
Crop 3531.9 4071.6 2784.9 3207.6
Crop (keep_size) 2581.0 2794.1 1480.6 1501.1
Pad 2220.0 2418.7 1749.7 1949.3
Pad (edge) 2210.9 2425.1 1721.7 1902.5
Pad (keep_size) 1789.2 1880.7 1042.6 1049.9
PadToFixedSize 2155.9 14188.0 1858.8 6742.3
CropToFixedSize 4144.2 41138.4 3262.9 10346.0
KeepSizeByResize (CropToFixedSize(nearest)) 2172.6 11017.3 1520.4 3290.5
KeepSizeByResize (CropToFixedSize(linear)) 2098.0 9433.9 1224.3 2225.6
KeepSizeByResize (CropToFixedSize(cubic)) 1992.2 7956.1 975.9 1516.6
FastSnowyLandscape 1679.9 2883.6 447.7 500.0
Clouds 212.7 214.5 52.9 52.3
Fog 361.2 362.0 86.9 86.6
CloudLayer 353.2 354.2 86.2 86.3
Snowflakes 174.5 178.3 76.6 82.6
SnowflakesLayer 353.4 358.5 167.2 166.5

Heatmaps

Numbers below are for heatmaps on large images, i.e. 224x224x3. Smaller images were skipped for brevity. The heatmaps themselves can be small (64x64xN) or large (224x224xN), with N denoting the number of heatmaps per HeatmapsOnImage instance (i.e. the number of channels in the heatmaps array), for which below 1 and 5 are used. B=1 denotes a batch size of 1 , B=128 one of 128.

mbit/sec for 64x64x5 or 224x224x5 heatmaps on 224x224x3 images:

  64x64x5 on 224x224x3 224x224x5 on 224x224x3
Augmenter B=1 B=128 B=1 B=128
Sequential (2xNoop) 1811.6 6545.7 14317.0 20101.4
Sequential (2xNoop, random_order) 1290.8 6483.3 11412.4 20071.6
SomeOf (1-3, 3xNoop) 814.4 4438.9 8120.7 18040.1
SomeOf (1-3, 3xNoop, random_order) 734.9 4407.0 7510.9 18613.1
OneOf (3xNoop) 1200.8 4592.7 10832.6 18840.6
Sometimes (Noop) 1124.8 6425.9 10353.8 20819.6
WithChannels ([1,2], Noop) 1730.2 6506.7 13820.9 20800.1
Noop 2006.4 6592.0 14986.0 20872.7
Lambda (return input) 1926.7 6544.8 14780.5 20888.9
AssertLambda (return True) 1925.7 6527.8 14777.5 20936.7
AssertShape (None, H, W, None) 1822.6 6321.0 14228.5 20714.5
ChannelShuffle (0.5) 2005.9 6571.7 14996.3 20884.7
Add 1999.1 6546.0 15010.3 20897.2
AddElementwise 1985.0 6572.4 14994.6 20910.0
AdditiveGaussianNoise 2005.5 6544.4 15006.6 20919.6
AdditiveLaplaceNoise 2002.2 6546.6 15007.0 20899.2
AdditivePoissonNoise 2003.5 6545.2 15037.2 20908.5
Multiply 1990.3 6572.0 15011.0 20925.0
MultiplyElementwise 2000.5 6532.7 14990.5 20946.8
Dropout (1-5%) 2000.1 6568.4 15002.4 20957.7
CoarseDropout (1-5%, size=1-10%) 1998.8 6579.9 14979.5 20849.6
ReplaceElementwise 1993.7 6582.4 15063.5 20856.3
ImpulseNoise 2006.6 6587.6 15017.4 20873.1
SaltAndPepper 1995.9 6575.9 15051.1 20869.2
CoarseSaltAndPepper 1994.3 6580.8 14983.9 20875.9
Salt 1992.8 6570.6 15079.1 20887.6
CoarseSalt 1986.9 6558.7 14965.5 20938.0
Pepper 1998.6 6525.3 14978.7 20907.4
CoarsePepper 2007.1 6513.3 15004.5 20873.0
Invert (10%) 2001.0 6540.9 15036.9 20875.8
JpegCompression (50-99%) 2004.2 6542.0 14930.0 20863.5
Alpha (Noop) 842.8 3055.5 6924.0 12675.7
AlphaElementwise (Noop) 216.9 279.7 1151.0 1228.0
SimplexNoiseAlpha (Noop) 94.2 102.5 764.1 794.5
FrequencyNoiseAlpha (Noop) 108.3 121.8 810.8 837.5
GaussianBlur (sigma=(1,5)) 1995.2 6549.6 15021.8 26701.2
AverageBlur 1997.9 6563.8 14984.2 26714.6
MedianBlur 2010.0 6547.5 15021.5 26712.7
BilateralBlur 2009.0 6539.5 14965.5 26682.9
MotionBlur 2004.0 6567.3 14914.4 26724.7
WithColorspace (HSV, Noop) 1797.1 6522.3 14165.3 26601.6
WithHueAndSaturation 1803.6 6506.9 14155.9 26595.1
MultiplyHueAndSaturation 1786.8 6513.5 13256.6 26554.6
MultiplyHue 1753.9 6529.9 13036.3 26620.9
MultiplySaturation 1745.1 6521.2 13044.0 26576.9
AddToHueAndSaturation 2002.2 6591.6 14707.4 26721.1
AddToHue 2004.7 6584.2 15035.1 26688.6
AddToSaturation 1996.1 6558.1 15087.1 26724.8
ChangeColorspace (HSV) 2006.0 6544.7 15076.4 26702.7
Grayscale 1999.8 6555.8 15147.1 26725.4
KMeansColorQuantization (2-16 colors) 2025.4 6560.4 15069.4 26736.8
UniformColorQuantization (2-16 colors) 2002.5 6566.2 15089.7 26666.8
GammaContrast 2005.9 6575.9 15006.6 26723.7
SigmoidContrast 2014.0 6596.6 14980.7 26683.9
LogContrast 2010.4 6570.3 15075.3 26694.8
LinearContrast 2007.0 6564.9 15126.8 26688.2
AllChannelsHistogramEqualization 2019.7 6550.6 15050.0 26611.4
HistogramEqualization 2118.0 6541.2 15044.9 26670.1
AllChannelsCLAHE 1999.5 6546.9 15077.3 26640.5
CLAHE 2001.9 6557.7 15246.5 26619.7
Convolve (3x3) 2118.1 6529.0 15005.7 26729.1
Sharpen 2116.3 6578.0 14990.8 26752.8
Emboss 2024.0 6586.0 14954.4 26737.4
EdgeDetect 2010.2 6591.2 15027.1 27517.4
DirectedEdgeDetect 2020.4 6589.9 15033.9 27775.1
Canny 2028.8 6588.8 15085.6 27780.6
Fliplr (p=100%) 1423.2 6095.4 12118.5 26993.1
Flipud (p=100%) 1455.5 6372.2 12312.4 27415.5
Affine (order=0, constant) 295.7 665.8 1168.9 1408.8
Affine (order=1, constant) 293.8 665.6 1172.2 1407.6
Affine (order=3, constant) 294.0 665.7 1173.7 1406.2
Affine (order=1, edge) 294.2 663.3 1176.6 1408.0
Affine (order=1, constant, skimage) 169.4 255.5 374.2 400.7
PiecewiseAffine (4x4, order=1, constant) 21.9 22.5 52.9 52.8
PiecewiseAffine (4x4, order=0, constant) 21.7 22.6 52.7 53.2
PiecewiseAffine (4x4, order=1, edge) 21.8 22.5 52.8 53.3
PiecewiseAffine (8x8, order=1, constant) 6.5 6.7 33.8 34.4
PerspectiveTransform 289.2 554.3 1496.4 1822.8
PerspectiveTransform (keep_size) 317.4 558.4 1270.5 1442.6
ElasticTransformation (order=0, constant) 85.6 101.1 1279.6 1506.7
ElasticTransformation (order=1, constant) 85.7 101.7 1279.4 1512.7
ElasticTransformation (order=1, nearest) 85.8 100.5 1276.9 1514.4
ElasticTransformation (order=1, reflect) 85.9 97.7 1279.1 1518.3
Rot90 958.4 5192.0 8719.2 25360.5
Rot90 (keep_size) 678.2 1903.0 4427.4 6720.6
AveragePooling 2013.5 6614.1 14378.3 20786.4
AveragePooling (keep_size) 2008.3 6611.3 14285.2 20785.6
MaxPooling 2010.8 6620.7 14377.5 20831.0
MaxPooling (keep_size) 1998.8 6632.3 14292.3 20897.1
MinPooling 1999.2 6632.3 14382.4 20898.9
MinPooling (keep_size) 2000.9 6632.6 14316.7 20893.2
MedianPooling 2011.0 6627.9 14319.1 20878.3
MedianPooling (keep_size) 2013.9 6629.7 14347.0 20838.3
Superpixels (max_size=64, cubic) 2013.0 6657.9 14412.7 20933.1
Superpixels (max_size=64, linear) 2013.6 6670.9 14474.7 20767.0
Superpixels (max_size=128, linear) 2010.7 6623.2 14484.4 20932.4
Superpixels (max_size=224, linear) 2008.8 6645.7 14433.6 20778.5
UniformVoronoi (250-1000k points, linear) 2010.2 6623.9 14397.1 20881.2
RegularGridVoronoi (16-31 rows/cols) 2002.7 6633.8 14186.9 20829.0
RelativeRegularGridVoronoi (7%-14% rows/cols) 1996.0 6581.1 14244.8 20939.4
Resize (nearest) 599.9 1414.1 4050.8 5859.1
Resize (linear) 586.1 1348.9 3614.8 5186.8
Resize (cubic) 571.3 1275.2 3196.3 4354.2
CropAndPad 541.6 876.0 3611.8 4618.5
CropAndPad (edge) 543.8 873.3 3587.6 4622.6
CropAndPad (keep_size) 405.2 588.1 2142.2 2453.9
Crop 742.3 1396.0 6703.0 8790.2
Crop (keep_size) 503.2 795.6 3211.1 3738.1
Pad 522.3 836.6 3004.5 3631.0
Pad (edge) 521.2 837.3 3012.9 3627.3
Pad (keep_size) 391.9 564.3 1859.8 2001.4
PadToFixedSize 527.8 1775.4 3131.2 5578.1
CropToFixedSize 788.6 3006.1 6630.7 14438.7
KeepSizeByResize (CropToFixedSize(nearest)) 470.3 1295.1 3214.6 4995.3
KeepSizeByResize (CropToFixedSize(linear)) 468.9 1249.5 2986.4 4515.5
KeepSizeByResize (CropToFixedSize(cubic)) 458.2 1182.5 2690.6 3862.0
FastSnowyLandscape 1994.5 6602.0 14273.7 26590.7
Clouds 825.0 4406.1 7776.7 22857.5
Fog 2012.6 6548.7 14466.8 26602.8
CloudLayer 2000.5 6555.3 14257.2 26598.1
Snowflakes 817.1 4399.1 7732.0 22826.1
SnowflakesLayer 2008.3 6552.2 14323.4 26585.7

Number of heatmap instances per sec for 64x64x5 or 224x224x5 heatmaps on 224x224x3 images:

  64x64x5 on 224x224x3 224x224x5 on 224x224x3
Augmenter B=1 B=128 B=1 B=128
Sequential (2xNoop) 14492.5 52365.2 9349.9 13127.4
Sequential (2xNoop, random_order) 10326.6 51866.5 7453.0 13108.0
SomeOf (1-3, 3xNoop) 6515.1 35511.3 5303.3 11781.3
SomeOf (1-3, 3xNoop, random_order) 5879.3 35256.3 4905.1 12155.5
OneOf (3xNoop) 9606.2 36741.3 7074.3 12304.0
Sometimes (Noop) 8998.3 51407.2 6761.7 13596.5
WithChannels ([1,2], Noop) 13841.9 52053.5 9025.9 13583.7
Noop 16051.4 52735.9 9786.8 13631.1
Lambda (return input) 15413.6 52358.2 9652.6 13641.7
AssertLambda (return True) 15405.2 52222.0 9650.6 13672.9
AssertShape (None, H, W, None) 14581.0 50567.8 9292.1 13527.8
ChannelShuffle (0.5) 16047.2 52573.7 9793.5 13639.0
Add 15992.5 52367.9 9802.6 13647.2
AddElementwise 15880.3 52579.4 9792.4 13655.5
AdditiveGaussianNoise 16044.4 52355.3 9800.2 13661.8
AdditiveLaplaceNoise 16017.8 52372.8 9800.5 13648.5
AdditivePoissonNoise 16027.9 52361.3 9820.2 13654.5
Multiply 15922.2 52575.8 9803.1 13665.3
MultiplyElementwise 16004.3 52261.8 9789.7 13679.5
Dropout (1-5%) 16000.6 52547.4 9797.5 13686.7
CoarseDropout (1-5%, size=1-10%) 15990.8 52639.1 9782.5 13616.1
ReplaceElementwise 15949.5 52659.1 9837.4 13620.5
ImpulseNoise 16052.9 52700.5 9807.3 13631.4
SaltAndPepper 15966.8 52607.1 9829.3 13628.9
CoarseSaltAndPepper 15954.6 52646.4 9785.4 13633.2
Salt 15942.2 52564.5 9847.6 13640.9
CoarseSalt 15895.4 52469.7 9773.4 13673.8
Pepper 15989.1 52202.4 9782.0 13653.8
CoarsePepper 16056.6 52106.7 9798.9 13631.4
Invert (10%) 16007.9 52327.4 9820.0 13633.2
JpegCompression (50-99%) 16033.4 52335.7 9750.2 13625.1
Alpha (Noop) 6742.3 24444.1 4521.8 8278.0
AlphaElementwise (Noop) 1735.5 2237.3 751.7 801.9
SimplexNoiseAlpha (Noop) 753.3 820.3 499.0 518.9
FrequencyNoiseAlpha (Noop) 866.5 974.5 529.5 546.9
GaussianBlur (sigma=(1,5)) 15961.3 52396.9 9810.1 17437.5
AverageBlur 15983.1 52510.6 9785.6 17446.3
MedianBlur 16080.3 52379.9 9809.9 17445.0
BilateralBlur 16072.3 52315.9 9773.4 17425.6
MotionBlur 16031.8 52538.4 9740.0 17452.9
WithColorspace (HSV, Noop) 14376.8 52178.3 9250.8 17372.5
WithHueAndSaturation 14428.5 52055.2 9244.7 17368.2
MultiplyHueAndSaturation 14294.4 52107.7 8657.4 17341.8
MultiplyHue 14031.0 52239.0 8513.5 17385.1
MultiplySaturation 13960.5 52169.9 8518.5 17356.3
AddToHueAndSaturation 16017.9 52732.8 9604.8 17450.5
AddToHue 16037.9 52673.3 9818.9 17429.3
AddToSaturation 15969.1 52464.7 9852.8 17453.0
ChangeColorspace (HSV) 16047.9 52357.9 9845.8 17438.5
Grayscale 15998.7 52446.6 9892.0 17453.3
KMeansColorQuantization (2-16 colors) 16203.6 52482.9 9841.2 17460.7
UniformColorQuantization (2-16 colors) 16020.1 52529.7 9854.5 17415.0
GammaContrast 16047.1 52607.0 9800.2 17452.2
SigmoidContrast 16111.9 52772.8 9783.3 17426.2
LogContrast 16082.8 52562.6 9845.1 17433.3
LinearContrast 16055.7 52519.5 9878.7 17429.0
AllChannelsHistogramEqualization 16157.8 52404.9 9828.6 17378.9
HistogramEqualization 16944.0 52329.6 9825.3 17417.2
AllChannelsCLAHE 15995.7 52375.2 9846.4 17397.9
CLAHE 16015.5 52461.5 9956.9 17384.3
Convolve (3x3) 16944.7 52232.4 9799.6 17455.7
Sharpen 16930.6 52624.0 9789.9 17471.2
Emboss 16192.1 52688.1 9766.1 17461.2
EdgeDetect 16081.3 52730.0 9813.6 17970.5
DirectedEdgeDetect 16162.9 52719.2 9818.0 18138.8
Canny 16230.8 52710.4 9851.8 18142.4
Fliplr (p=100%) 11386.0 48763.0 7914.1 17628.2
Flipud (p=100%) 11644.2 50977.6 8040.8 17904.0
Affine (order=0, constant) 2365.5 5326.1 763.3 920.0
Affine (order=1, constant) 2350.2 5324.7 765.5 919.3
Affine (order=3, constant) 2352.3 5325.9 766.5 918.3
Affine (order=1, edge) 2353.7 5306.7 768.4 919.5
Affine (order=1, constant, skimage) 1355.4 2044.0 244.4 261.7
PiecewiseAffine (4x4, order=1, constant) 175.4 180.3 34.5 34.5
PiecewiseAffine (4x4, order=0, constant) 173.8 180.5 34.4 34.8
PiecewiseAffine (4x4, order=1, edge) 174.1 180.1 34.5 34.8
PiecewiseAffine (8x8, order=1, constant) 52.3 53.7 22.1 22.4
PerspectiveTransform 2313.6 4434.0 977.2 1190.4
PerspectiveTransform (keep_size) 2539.1 4467.3 829.7 942.1
ElasticTransformation (order=0, constant) 684.6 809.0 835.6 984.0
ElasticTransformation (order=1, constant) 685.3 813.7 835.5 987.9
ElasticTransformation (order=1, nearest) 686.7 803.7 833.9 989.0
ElasticTransformation (order=1, reflect) 686.9 781.7 835.4 991.6
Rot90 7667.3 41535.9 5694.2 16561.9
Rot90 (keep_size) 5425.4 15223.7 2891.3 4388.9
AveragePooling 16108.0 52912.6 9389.9 13574.8
AveragePooling (keep_size) 16066.1 52890.2 9329.1 13574.2
MaxPooling 16086.2 52965.4 9389.4 13603.9
MaxPooling (keep_size) 15990.2 53058.5 9333.7 13647.1
MinPooling 15993.5 53058.5 9392.6 13648.3
MinPooling (keep_size) 16006.9 53060.4 9349.7 13644.6
MedianPooling 16087.7 53023.6 9351.2 13634.8
MedianPooling (keep_size) 16110.8 53037.3 9369.5 13608.7
Superpixels (max_size=64, cubic) 16104.4 53263.3 9412.4 13670.6
Superpixels (max_size=64, linear) 16109.0 53367.3 9452.9 13562.1
Superpixels (max_size=128, linear) 16085.2 52985.9 9459.2 13670.1
Superpixels (max_size=224, linear) 16070.1 53165.8 9426.0 13569.6
UniformVoronoi (250-1000k points, linear) 16081.6 52991.4 9402.2 13636.7
RegularGridVoronoi (16-31 rows/cols) 16021.4 53070.2 9264.9 13602.6
RelativeRegularGridVoronoi (7%-14% rows/cols) 15968.0 52649.0 9302.7 13674.7
Resize (nearest) 4799.3 11313.0 2645.4 3826.3
Resize (linear) 4689.1 10791.3 2360.7 3387.3
Resize (cubic) 4570.4 10201.3 2087.4 2843.6
CropAndPad 4332.5 7008.1 2358.8 3016.2
CropAndPad (edge) 4350.3 6986.6 2342.9 3018.9
CropAndPad (keep_size) 3241.6 4704.6 1399.0 1602.6
Crop 5938.7 11168.3 4377.5 5740.5
Crop (keep_size) 4025.6 6364.8 2097.0 2441.2
Pad 4178.2 6692.7 1962.1 2371.3
Pad (edge) 4169.9 6698.1 1967.6 2368.8
Pad (keep_size) 3135.3 4514.0 1214.6 1307.0
PadToFixedSize 4222.2 14203.1 2044.9 3642.8
CropToFixedSize 6308.6 24048.9 4330.3 9429.4
KeepSizeByResize (CropToFixedSize(nearest)) 3762.0 10360.8 2099.3 3262.2
KeepSizeByResize (CropToFixedSize(linear)) 3751.4 9996.4 1950.3 2948.9
KeepSizeByResize (CropToFixedSize(cubic)) 3665.4 9459.8 1757.1 2522.1
FastSnowyLandscape 15956.0 52816.3 9321.6 17365.4
Clouds 6599.7 35248.4 5078.7 14927.4
Fog 16100.6 52389.4 9447.7 17373.3
CloudLayer 16003.9 52442.6 9310.9 17370.2
Snowflakes 6536.7 35192.9 5049.5 14906.8
SnowflakesLayer 16066.5 52417.9 9354.0 17362.1

Keypoints and Bounding Boxes

Numbers below are for keypoints on small and large images. Each KeypointsOnImage instance contained 10 Keypoint instances. B=1 denotes a batch size of 1 , B=128 one of 128.

The numbers for bounding boxes can be derived by dividing each value by 4.

Number of augmented Keypoint instances per sec (divide by 10 for KeypointsOnImage instances):

  10 KPs on 64x64x3 10 KPs on 224x224x3
Augmenter B=1 B=128 B=1 B=128
Sequential (2xNoop) 117845.1 1271754.9 116128.9 1254123.2
Sequential (2xNoop, random_order) 38948.3 1078527.6 39444.1 1063884.6
SomeOf (1-3, 3xNoop) 17187.8 182412.7 17412.7 185448.5
SomeOf (1-3, 3xNoop, random_order) 15297.5 180090.2 15496.1 181975.2
OneOf (3xNoop) 33475.4 201685.5 34231.4 204038.2
Sometimes (Noop) 27682.4 940970.8 28066.7 931063.2
WithChannels ([1,2], Noop) 97010.3 1140444.1 96834.8 1117121.1
Noop 150981.1 1280647.5 152547.9 1251029.4
Lambda (return input) 142708.7 1240949.9 142966.5 1227180.3
AssertLambda (return True) 142859.4 1236418.2 143342.7 1222915.5
AssertShape (None, H, W, None) 106638.5 905189.5 108450.9 905424.6
ChannelShuffle (0.5) 151875.8 1256490.9 154113.6 1243465.8
Add 152366.9 1266360.3 154145.7 1241734.4
AddElementwise 152018.9 1265638.8 153296.8 1241544.9
AdditiveGaussianNoise 152221.2 1257915.8 153586.9 1245358.0
AdditiveLaplaceNoise 152344.7 1256556.6 152549.7 1246072.9
AdditivePoissonNoise 150791.1 1266635.2 152699.7 1234008.2
Multiply 151215.1 1269893.9 153156.9 1244439.0
MultiplyElementwise 151501.0 1274457.9 152840.6 1248975.6
Dropout (1-5%) 150917.7 1270715.4 154313.9 1244631.4
CoarseDropout (1-5%, size=1-10%) 152883.4 1271874.4 153311.8 1241941.2
ReplaceElementwise 152256.2 1263005.8 152812.8 1249882.8
ImpulseNoise 151877.7 1261434.9 153015.4 1243879.7
SaltAndPepper 153000.5 1261012.2 152155.0 1249304.0
CoarseSaltAndPepper 151226.0 1259696.5 153405.2 1243867.2
Salt 152177.1 1262169.4 152970.7 1242126.1
CoarseSalt 150292.2 1261297.6 149935.8 1244189.1
Pepper 153444.5 1261691.9 150211.4 1225298.1
CoarsePepper 152280.2 1259434.5 151182.4 1230634.6
Invert (10%) 153145.7 1264599.4 152459.2 1222115.6
JpegCompression (50-99%) 152236.0 1270393.7 152342.9 1222478.3
Alpha (Noop) 23645.3 405669.6 23982.6 398329.0
AlphaElementwise (Noop) 12494.4 22371.1 7059.9 9582.6
SimplexNoiseAlpha (Noop) 2703.5 3065.8 1840.6 2044.0
FrequencyNoiseAlpha (Noop) 3129.0 3814.4 2198.9 2552.7
GaussianBlur (sigma=(1,5)) 151802.5 1242964.9 154881.9 1238279.4
AverageBlur 150304.7 1253134.7 151146.1 1238891.9
MedianBlur 152096.1 1257904.0 151599.5 1252962.2
BilateralBlur 152809.1 1260036.5 150718.8 1239920.0
MotionBlur 153229.6 1264048.5 151687.2 1249910.0
WithColorspace (HSV, Noop) 109771.7 1211901.0 109518.5 1189720.5
WithHueAndSaturation 110064.6 1208509.1 110056.9 1188839.7
MultiplyHueAndSaturation 110358.0 1199819.7 108910.0 1189365.6
MultiplyHue 100584.4 1143882.4 99356.6 1131482.9
MultiplySaturation 100817.3 1143674.5 99957.2 1126901.5
AddToHueAndSaturation 153052.6 1251988.2 151059.0 1232845.5
AddToHue 151903.3 1252179.0 152361.3 1236216.1
AddToSaturation 153332.3 1255361.7 151322.4 1231305.4
ChangeColorspace (HSV) 153737.0 1260859.2 153283.8 1231802.6
Grayscale 153703.2 1259484.8 152053.8 1236252.1
KMeansColorQuantization (2-16 colors) 155598.2 1260118.4 150182.8 1226729.7
UniformColorQuantization (2-16 colors) 153718.2 1253918.2 151320.6 1235727.6
GammaContrast 155114.8 1258490.8 150885.1 1235241.4
SigmoidContrast 154779.0 1263570.6 150888.7 1239783.6
LogContrast 154281.8 1263292.1 151375.2 1231757.4
LinearContrast 153508.2 1265834.8 151062.6 1226712.0
AllChannelsHistogramEqualization 150080.7 1253029.5 145686.1 1232670.9
HistogramEqualization 153661.9 1251161.5 150362.2 1245596.8
AllChannelsCLAHE 153465.1 1259210.0 150306.5 1238668.9
CLAHE 156022.6 1267221.1 151630.6 1234053.6
Convolve (3x3) 154329.1 1258661.9 147537.8 1237043.1
Sharpen 154140.0 1262945.4 151269.6 1242612.9
Emboss 153480.1 1261828.3 151669.0 1231888.4
EdgeDetect 151433.5 1266341.4 150504.3 1229935.4
DirectedEdgeDetect 151453.5 1264529.9 151118.9 1234051.7
Canny 152562.7 1261503.1 152644.1 1232776.6
Fliplr (p=100%) 41840.8 767543.2 42707.5 759180.8
Flipud (p=100%) 42311.4 759172.6 42454.8 757060.1
Affine (order=0, constant) 7878.6 35591.2 7909.1 36589.5
Affine (order=1, constant) 7829.1 35621.3 7913.4 36506.3
Affine (order=3, constant) 7826.7 35533.0 7912.8 36425.5
Affine (order=1, edge) 7781.4 35404.6 7900.8 36482.5
Affine (order=1, constant, skimage) 7835.9 35408.7 7941.2 36478.5
PiecewiseAffine (4x4, order=1, constant) 428.7 443.6 125.2 127.1
PiecewiseAffine (4x4, order=0, constant) 432.1 442.9 124.9 127.3
PiecewiseAffine (4x4, order=1, edge) 428.3 444.1 124.7 127.2
PiecewiseAffine (8x8, order=1, constant) 111.7 112.2 61.2 62.9
PerspectiveTransform 12129.5 25392.6 12296.1 25890.5
PerspectiveTransform (keep_size) 10071.7 18472.1 10195.8 18809.4
ElasticTransformation (order=0, constant) 1414.1 1668.3 1225.3 1584.4
ElasticTransformation (order=1, constant) 1615.8 1699.2 1415.4 1595.0
ElasticTransformation (order=1, nearest) 1316.7 1680.3 1402.2 1567.9
ElasticTransformation (order=1, reflect) 1448.0 1690.5 1472.5 1576.5
Rot90 20993.4 148342.8 23813.4 341765.4
Rot90 (keep_size) 20933.8 145783.9 23829.1 339986.8
AveragePooling 148330.9 1291550.6 147253.0 1266495.7
AveragePooling (keep_size) 148397.4 1292584.1 148131.9 1265094.0
MaxPooling 151398.9 1289432.0 150720.6 1262554.3
MaxPooling (keep_size) 150419.7 1293971.5 150753.1 1253855.7
MinPooling 151048.1 1295758.9 153239.0 1250585.4
MinPooling (keep_size) 150087.8 1297152.1 151253.3 1256454.6
MedianPooling 151568.5 1291223.5 152510.9 1257977.7
MedianPooling (keep_size) 149866.2 1292432.6 151905.2 1264669.9
Superpixels (max_size=64, cubic) 151120.7 1291059.9 151960.2 1264689.7
Superpixels (max_size=64, linear) 151273.3 1292381.8 151455.4 1273175.4
Superpixels (max_size=128, linear) 152195.5 1290679.2 153615.0 1270632.2
Superpixels (max_size=224, linear) 150319.1 1291764.0 154444.6 1272684.5
UniformVoronoi (250-1000k points, linear) 151222.4 1287975.0 151055.4 1267464.5
RegularGridVoronoi (16-31 rows/cols) 146866.2 1290315.2 151747.6 1272045.2
RelativeRegularGridVoronoi (7%-14% rows/cols) 150527.7 1293253.5 152772.0 1267959.4
Resize (nearest) 17175.4 52737.9 17414.3 52118.1
Resize (linear) 17243.3 52844.3 17381.1 52075.5
Resize (cubic) 17180.8 52668.0 17451.6 51909.1
CropAndPad 13521.4 22962.3 15101.4 27471.4
CropAndPad (edge) 13488.6 22848.6 15046.7 27381.1
CropAndPad (keep_size) 11022.8 16300.5 11953.3 18389.7
Crop 16725.9 30934.7 18981.1 39310.5
Crop (keep_size) 12868.3 20043.9 14234.8 22841.6
Pad 13565.9 23164.6 15140.9 27644.5
Pad (edge) 13603.5 23199.1 15087.0 27581.4
Pad (keep_size) 10961.7 16365.4 11954.7 18546.6
PadToFixedSize 13644.6 128647.2 15016.8 294885.8
CropToFixedSize 18874.6 129670.2 21340.0 306892.3
KeepSizeByResize (CropToFixedSize(nearest)) 11664.4 46163.9 12589.2 57793.6
KeepSizeByResize (CropToFixedSize(linear)) 11611.3 45703.5 12591.7 57628.1
KeepSizeByResize (CropToFixedSize(cubic)) 11649.4 45623.2 12575.0 56999.1
FastSnowyLandscape 155068.9 1295390.0 152534.9 1270551.0
Clouds 17421.8 188714.3 17714.3 190750.6
Fog 153315.5 1270560.0 152950.3 1263348.5
CloudLayer 155061.3 1278450.8 153938.2 1265986.0
Snowflakes 17105.7 186278.8 17504.0 188652.1
SnowflakesLayer 153196.1 1270147.2 154954.3 1252210.2