AlbumentationsX vs Kornia Benchmark

Use this page for measured throughput. The Summary page covers the library-level tradeoffs, and the Transform Mapping page covers capability alignment.

Benchmark vs Kornia

The benchmark data is generated during the site build from the published benchmark artifacts. Results are shown for the full PyTorch DataLoader pipeline and for the single-operation RGB micro benchmark. Published GPU artifacts are shown in the DataLoader table where they exist; the RGB micro table on this page is CPU.

DataLoader Pipeline

End-to-end in-memory DataLoader throughput with 8 workers, batch size 256, normalization, tensor conversion, batching, and collation included. The table shows the variable transform name only. Higher images/second is better. CPU and GPU columns are shown where published artifacts exist.

56 / 57
pipelines where AlbumentationsX is faster
3.6x
average speedup
average computed from benchmark table rows
2.2.6 vs 0.8.2
library versions
from generated benchmark metadata
TransformSpeedup
Albx / best
AlbumentationsX
CPU · 2.2.6
Kornia
CPU · 0.8.2
Kornia
GPU · 0.8.2
MedianBlur12x4038 ± 089 ± 0327 ± 2
Elastic9.0-9.2x2878 ± 0103 ± 0316 ± 2
JpegCompression5.5x4267 ± 0771 ± 0590 ± 2
ColorJiggle5.3x4255 ± 0800 ± 0529 ± 2
PlasmaBrightness4.7-5.1x2727 ± 0458 ± 0562 ± 22
CLAHE4.4x3505 ± 0789 ± 0165 ± 0
ColorJitter4.4x4299 ± 0987 ± 0573 ± 3
Blur4.2x5507 ± 01321 ± 0651 ± 2
Hue4.1x4723 ± 01144 ± 0586 ± 6
PlasmaContrast4.1x2291 ± 0455 ± 0557 ± 2
Saturation4.1x4646 ± 01141 ± 0587 ± 4
Sharpen3.9x4978 ± 01267 ± 0627 ± 5
GaussianBlur3.9x5029 ± 01289 ± 0650 ± 5
MotionBlur3.9x4785 ± 01238 ± 0657 ± 2
Snow3.8x4135 ± 01102 ± 0596 ± 3
Erasing3.6x5415 ± 01522 ± 0503 ± 2
Solarize3.4x5407 ± 01609 ± 0644 ± 2
RandomGamma3.4x5450 ± 01624 ± 0653 ± 5
Equalize3.4x4511 ± 01344 ± 0321 ± 0
RandomRotate903.3x5087 ± 01520 ± 0632 ± 8
Posterize3.3x5431 ± 01660 ± 0562 ± 3
Rotate3.3x4865 ± 01488 ± 0642 ± 5
ChannelShuffle3.1x5467 ± 01739 ± 0643 ± 2
Perspective3.1x4132 ± 01325 ± 0635 ± 1
SaltAndPepper3.1x4508 ± 01462 ± 0375 ± 11
ChannelDropout3.0x5314 ± 01742 ± 0660 ± 3
RandomResizedCrop3.0x5056 ± 01659 ± 0905 ± 7
Contrast3.0x5435 ± 01829 ± 0649 ± 5
Invert2.9x5569 ± 01898 ± 0659 ± 3
PlasmaShadow2.9x2797 ± 0956 ± 0612 ± 22
RandomJigsaw2.9x4863 ± 01664 ± 0636 ± 2
VerticalFlip2.9x5512 ± 01887 ± 0669 ± 2
Grayscale2.9x5378 ± 01845 ± 0655 ± 2
Brightness2.9x5310 ± 01832 ± 0651 ± 5
Affine2.9x4528 ± 01566 ± 0646 ± 2
PlankianJitter2.8x4979 ± 01763 ± 0654 ± 8
RGBShift2.8x4977 ± 01770 ± 0658 ± 3
RandomCrop2242.8x5130 ± 01856 ± 0959 ± 9
HorizontalFlip2.8x5002 ± 01819 ± 0670 ± 2
Rain2.7x4198 ± 01541 ± 0306 ± 2
AutoContrast2.7x4646 ± 01721 ± 0651 ± 4
Shear2.7x4141 ± 01557 ± 0
CornerIllumination2.6x3969 ± 01502 ± 0395 ± 2
GaussianIllumination2.5x3656 ± 01458 ± 0
Resize2.5x1350 ± 0538 ± 0181 ± 0
OpticalDistortion2.5x3579 ± 01445 ± 0641 ± 6
SmallestMaxSize2.5x1370 ± 0555 ± 0179 ± 0
LongestMaxSize2.4x1363 ± 0560 ± 0179 ± 0
LinearIllumination2.4x4179 ± 01726 ± 0503 ± 2
GaussianNoise2.1x3415 ± 01634 ± 0660 ± 7
ThinPlateSpline0.9x743 ± 0793 ± 0598 ± 2
EnhanceDetail5033 ± 0
EnhanceEdge4923 ± 0
Pad4996 ± 0
PhotoMetricDistort4269 ± 0
Transpose5338 ± 0
UnsharpMask4635 ± 0

Micro Benchmark

The benchmark below compares single-threaded RGB image augmentation throughput. Kornia receives normalized tensors; AlbumentationsX receives OpenCV-loaded RGB NumPy arrays. Higher images/second is better.

51 / 51
transforms where AlbumentationsX is faster
9.10x
median speedup
4.61x-17.06x IQR
2.2.6 vs 0.8.2
library versions
from generated benchmark metadata
TransformSpeedup
Albx / best
albumentationsx
2.2.6
Kornia
0.8.2
Blur78-79x4449 ± 1757 ± 0
Posterize48-51x14399 ± 58290 ± 9
Solarize45-46x9760 ± 34214 ± 1
MedianBlur≥42x843 ± 4≤20
GaussianBlur41x2343 ± 457 ± 0
RandomCrop22439-40x38380 ± 192981 ± 5
RandomGamma32-33x9938 ± 46308 ± 2
Erasing32x9511 ± 74298 ± 1
MotionBlur24-25x1953 ± 2181 ± 1
Sharpen24x1388 ± 558 ± 0
RandomJigsaw23-24x5172 ± 16219 ± 2
ColorJiggle19x639 ± 534 ± 0
RandomRotate9018x5990 ± 85333 ± 4
JpegCompression16x692 ± 743 ± 0
Invert15x15095 ± 611015 ± 2
Hue15x967 ± 166 ± 0
PlasmaBrightness≥13x267 ± 1≤20
VerticalFlip13x14051 ± 551067 ± 2
Saturation12-13x847 ± 1767 ± 0
Grayscale12x5194 ± 1418 ± 1
ColorJitter12-13x641 ± 152 ± 1
RandomResizedCrop11-12x7150 ± 19622 ± 3
Elastic≥9.5x191 ± 0≤20
SmallestMaxSize9.3-9.6x2017 ± 25214 ± 1
HorizontalFlip9.0-9.3x8416 ± 19920 ± 10
Resize8.9-9.3x2463 ± 37271 ± 1
Brightness8.9-9.1x6912 ± 13766 ± 7
Contrast8.8-9.1x6933 ± 30771 ± 9
ChannelShuffle8.9-9.0x4337 ± 13487 ± 2
LongestMaxSize8.4-8.7x2825 ± 42330 ± 1
ChannelDropout8.1-8.4x6810 ± 65828 ± 6
PlasmaShadow7.9-8.0x420 ± 353 ± 0
Snow7.8-7.9x489 ± 362 ± 0
PlasmaContrast≥7.1x143 ± 0≤20
Equalize6.3x807 ± 3128 ± 0
AutoContrast5.3-5.5x1243 ± 19231 ± 1
SaltAndPepper4.7-4.9x738 ± 10154 ± 1
GaussianNoise4.6x225 ± 049 ± 0
CLAHE4.6x283 ± 162 ± 0
Rotate4.2-4.5x1408 ± 40325 ± 2
PlankianJitter3.8-3.9x2253 ± 17580 ± 2
RGBShift3.8-3.9x2292 ± 3597 ± 3
Perspective3.1x559 ± 2181 ± 1
CornerIllumination2.7x425 ± 2157 ± 0
Rain2.4x1259 ± 2527 ± 5
Affine2.1-2.2x872 ± 8402 ± 3
GaussianIllumination2.1x388 ± 1188 ± 0
Shear1.9-2.0x784 ± 6403 ± 1
LinearIllumination1.6x521 ± 1327 ± 3
ThinPlateSpline1.4x52 ± 036 ± 0
OpticalDistortion1.4x274 ± 1201 ± 1
EnhanceDetail2148 ± 13
EnhanceEdge1373 ± 16
Pad13181 ± 118
PhotoMetricDistort581 ± 4
Transpose4627 ± 26
UnsharpMask906 ± 2

See the aggregate image benchmark or inspect the benchmark source code.

Related Pages