enot.ai - benchmarks

Benchmarks

This section features benchmarks for various widely used open-source neural network models. In each case a baseline benchmark was run to establish the initial accuracy and performance. Then, each underlying model was optimized by using enot.ai’s Neural network architecture selection engine, after which the resulting optimized version of the model was inferred on enot.ai’s custom inference engine to obtain the resulting benchmar

Select model:

Yolo-v5s

MobileNetv2

Mobilenet v2SSD

Resnet50

Resnet18

BERT

Other models

Choose hardware*:

GPUs

Edge CPUs

Server CPUs

Select dataset:

CIFAR-100

Mseg

Imagenette

Pascal VOC

XYZ

Other datasets

Yolo_v5s
(COCO)

mAp

FPS

Acceleration

Baseline

0.43

148

1.0x

enot.ai_optimized

0.42

1021

6.9x

Resnet-50
(CIFAR-100)

Accuracy

MMAC

Acceleration

Baseline

0.76

170

1.0x

enot.ai_optimized

0.75

1178

6.9x

Resnet-50
(CIFAR-100)

Accuracy, mAp

FPS

Acceleration

Baseline

0.76

49

1.0x

enot.ai_optimized

0.75

406

8.3x

Resnet-50
(Imagenette)

Accuracy, mAp

FPS

Acceleration

Baseline

0.89

190

1.0x

enot.ai_optimized

0.88

2128

11.2x

Resnet-18
(CIFAR-100)

Accuracy

MMAC

Acceleration

Baseline

0.74

547

1.0x

enot.ai_optimized

0.73

92

5.9x

Resnet-18
(CIFAR-100)

Accuracy

FPS

Acceleration

Baseline

0.74

155

1.0x

enot.ai_optimized

0.73

1272

8.2x

MobileNet_v2
(Imagenette)

Accuracy

FPS

Acceleration

Baseline

0.89

190

1.0x

enot.ai_optimized

0.88

2128

11.2x

MobileNet_v2
(Imagenette)

Accuracy

MMAC

Acceleration

Baseline

0.89

299

1.0x

enot.ai_optimized

0.88

75

4.0x

MobileNet_v2
(Imagenette)

Accuracy

FPS

Acceleration

Baseline

0.89

3.6

1.0x

enot.ai_optimized

0.88

31.4

8.7x

MobileNet_v2
(CIFAR-100)

Accuracy

MMAC

Acceleration

Baseline

0.68

54.4

1.0x

enot.ai_optimized

0.67

17.3

3.1x

MobileNet_v2
(CIFAR-100)

Accuracy

FPS

Acceleration

Baseline

0.68

180

1.0x

enot.ai_optimized

0.67

1598

8.9x

MobileNet v2SSD
(Pascal VOC)

mAP

FPS

Acceleration

Baseline

0.41

124

1.0x

enot.ai_optimized

0.4

1537

0.1x

MobileNet v2SSD
(Pascal VOC)

mAP

FPS

Acceleration

Baseline

0.41

2

1.0x

enot.ai_optimized

0.4

13.8

6.9x

MobileNet v2SSD
(Pascal VOC)

mAP

MMAC

Acceleration

Baseline

0.65

1340

1.0x

enot.ai_optimized

0.63

527

2.5x

HR-net
(Mseg)

MIoU

MMAC

Acceleration

Baseline

0.44

46256

1.0x

enot.ai_optimized

0.43

17731

2.6x

HR-net
(Mseg)

MIoU

FPS

Acceleration

Baseline

0.44

11

1.0x

enot.ai_optimized

0.43

47.2

4.3x

BERT
(TBD)

F1

FPS

Acceleration

Baseline

-

-

1.0x

enot.ai_optimized

90.5

846

9.3x

Vision Transformer (ViT)
Mseg

MIoU

FPS

Acceleration

Baseline

-

-

1.0x

enot.ai_optimized

175

577

3.3x

*Hardware
Edge CPUs include ARM, NPUs, ARC, <1GHz clock speed
GPUs include Nvidia-based graphical processing units
Server CPUs include Intel-based CPUs