Skip to content
KernelIndex
Search⌘K
Sign in

Multi scale feature pyramid

25 eligible runs
vision

Multi-scale feature pyramid processing for hierarchical image processing with encoder-decoder architecture and skip connections. Implements U-Net style progressive downsampling and upsampling with channel expansion/reduction at each scale level.

Fastest reported
482.2µsmean · 64% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-15. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencynum_scales = 3 · base_channels = 32 · input_channels = 3 · scale1_channels = 32 · scale2_channels = 64 · scale3_channels = 12816 cases
height
width
batch_size
dtype
64644fp3264648fp32646416fp321281281fp321281284fp3212812832fp3212812864fp321321321fp322562562fp322562568fp3225625616fp323723728fp324484481fp325125121fp325125122fp32102410241fp32
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 1482.2 µsRIAC as well19NVIDIA B200 · env 2547.8 µsdoubleAI6

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload suite of 16 cases · mean latency · Protocol SOL-ExecBench evaluation stack v1.1 · mean · 19 results · last observed 2026-08-24Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
RIAC as well
482.2µs64% SOL
1.00×
Reported · license unknown · no source
2026-08-15

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
2
doubleAI
554.4µs61% SOL
1.15×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
3
Amir M. Mir | SF Tensor
609.7µs59% SOL
1.26×
Reported · license unknown · no source
2026-07-20

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
4
ac4k
657.1µs57% SOL
1.36×
Reported · license unknown · no source
2026-08-04

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
5
Hyra
693.9µs55% SOL
1.44×
Reported · license unknown · no source
2026-07-19

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
6
Recursive
752.2µs53% SOL
1.56×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
7
Geometric
804.7µs52% SOL
1.67×
Reported · license unknown · no source
2026-08-23

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
8
Geometric
805.4µs52% SOL
1.67×
Reported · license unknown · no source
2026-08-22

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
9
Geometric
808.1µs52% SOL
1.68×
Reported · license unknown · no source
2026-08-24

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
10
Geometric
855.4µs50% SOL
1.77×
Reported · license unknown · no source
2026-08-15

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
11
Geometric
868.2µs50% SOL
1.80×
Reported · license unknown · no source
2026-08-21

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
12
Geometric
904.7µs49% SOL
1.88×
Reported · license unknown · no source
2026-08-16

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
13
Amir M. Mir | SF Tensor
908.3µs49% SOL
1.88×
Reported · license unknown · no source
2026-07-14

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
14
Geometric
933.1µs48% SOL
1.94×
Reported · license unknown · no source
2026-08-13

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
15
Geometric
995.5µs47% SOL
2.06×
Reported · license unknown · no source
2026-08-15

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
16
Amir M. Mir | SF Tensor
1.00ms46% SOL
2.08×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
17
Proud Wren @ NVIDIA
1.00ms46% SOL
2.08×
Reported · license unknown · no source
2026-07-11

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
18
gnehiz
1.03ms45% SOL
2.14×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
19
Geometric
1.05ms45% SOL
2.17×
Reported · license unknown · no source
2026-07-23

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
RIAC as well
unknown
482.2µs
1.00×
Reported
license unknown · no source
ac4k
unknown
657.1µs
1.36×
Reported
license unknown · no source
Hyra
unknown
693.9µs
1.44×
Reported
license unknown · no source
Recursive
unknown
733.5µs
1.52×
Reported
license unknown · no source
Geometric
unknown
804.7µs
1.67×
Reported
license unknown · no source
Cursor
unknown
845.6µs
1.75×
Reported
license unknown · no source
Proud Wren @ NVIDIA
unknown
1.00ms
2.08×
Reported
license unknown · no source
gnehiz
unknown
1.02ms
2.12×
Reported
license unknown · no source
Show all 25 implementations ›
Recursive
unknown
752.2µs
1.56×
Reported
license unknown · no source
Geometric
unknown
805.4µs
1.67×
Reported
license unknown · no source
Geometric
unknown
808.1µs
1.68×
Reported
license unknown · no source
Geometric
unknown
855.4µs
1.77×
Reported
license unknown · no source
Geometric
unknown
868.2µs
1.80×
Reported
license unknown · no source
doubleAI
unknown
870.1µs
1.80×
Reported
license unknown · no source
Geometric
unknown
904.7µs
1.88×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
908.3µs
1.88×
Reported
license unknown · no source
Geometric
unknown
933.1µs
1.94×
Reported
license unknown · no source
Geometric
unknown
995.5µs
2.06×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
991.2µs
2.06×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
1.00ms
2.08×
Reported
license unknown · no source
gnehiz
unknown
1.03ms
2.14×
Reported
license unknown · no source
Geometric
unknown
1.05ms
2.17×
Reported
license unknown · no source
doubleAI
unknown
547.8µs
1.14×
Reported
license unknown · no source
doubleAI
unknown
554.4µs
1.15×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
609.7µs
1.26×
Reported
license unknown · no source

Semantics

Inputs and outputs
xfp32 [batch_size, input_channels, height, width]
encoder_conv0_weightfp32 [32, 3, 3, 3]
encoder_conv0_biasfp32 [32]
downsample_conv0_weightfp32 [64, 32, 3, 3]
downsample_conv0_biasfp32 [64]
encoder_conv1_weightfp32 [64, 64, 3, 3]
encoder_conv1_biasfp32 [64]
downsample_conv1_weightfp32 [128, 64, 3, 3]
downsample_conv1_biasfp32 [128]
encoder_conv2_weightfp32 [128, 128, 3, 3]
encoder_conv2_biasfp32 [128]
bottleneck_weightfp32 [128, 128, 3, 3]
bottleneck_biasfp32 [128]
upsample_conv0_weightfp32 [128, 64, 2, 2]
upsample_conv0_biasfp32 [64]
decoder_conv0_weightfp32 [64, 128, 3, 3]
decoder_conv0_biasfp32 [64]
upsample_conv1_weightfp32 [64, 32, 2, 2]
upsample_conv1_biasfp32 [32]
decoder_conv1_weightfp32 [32, 64, 3, 3]
decoder_conv1_biasfp32 [32]
output_conv_weightfp32 [3, 32, 3, 3]
output_conv_biasfp32 [3]
outputfp32 [batch_size, input_channels, height, width]
Axes and behavior
widthvariable
heightvariable
batch_sizevariable
num_scalesconstant = 3
width_div2width_div2 = width // 2
width_div4width_div4 = width // 4
height_div2height_div2 = height // 2
height_div4height_div4 = height // 4
base_channelsconstant = 32
input_channelsconstant = 3
scale1_channelsscale1_channels = base_channels
scale2_channelsscale2_channels = base_channels * 2
scale3_channelsscale3_channels = base_channels * 4
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias033_multi_scale_feature_pyramidmodelnafnetsha256a5c320b3e5fd…
Sources: NVIDIA SOL-ExecBench (2026-08-24)last observed 2026-08-24How records are decidedJSON