Dashboard
Devices
Data acquisition
Experiments
- EON Tuner
Impulse design
Impulse #1
Versioning

Getting started

Documentation
Forums

Dhruv Sheth / cloud-classification-spresense Public

Clone this project

Dhruv Sheth / cloud-classification-spresense Public

Target: Sony Spresense (Cortex-M4F 156MHz) Clone this project

Experiments EON Tuner

The EON Tuner helps you find the most optimal architecture for your embedded machine-learning application. Clone this project to use the EON Tuner.

Target

No name set

Sony Spresense (Cortex-M4F 156MHz)

100 ms

1536 kB

8096 kB

Filters

Status

Pending 0

Running 3

Completed 42

Failed 5

DSP type

Grayscale 25

RGB 25

Model type

Convolutional (2D) 22

mobilenetv1 11

mobilenetv2 17

View

Data set

Validation

Test

Variant

Sort

General

Accuracy

Latency

RAM

ROM

Last updated

F1-score

Precision

Recall

96%

rgb-conv2d-6d6

PERFORMANCE

LATENCY

448 ms of 100 ms

Exceeds target by 348 ms

RAM

30 kB of 1536 kB

ROM

122 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

32 |

32

IMAGE

RGB

ACCURACY (KERAS)

12/15/2021, 2:39:58 PM

96%

grayscale-mobilenetv2-368

PERFORMANCE

LATENCY

6291 ms of 100 ms

Exceeds target by 6191 ms

RAM

351 kB of 1536 kB

ROM

585 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 1:01:57 PM

96%

grayscale-mobilenetv2-c78

PERFORMANCE

LATENCY

1825 ms of 100 ms

Exceeds target by 1725 ms

RAM

64 kB of 1536 kB

ROM

109 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 1:37:45 PM

96%

grayscale-mobilenetv2-4b9

PERFORMANCE

LATENCY

1825 ms of 100 ms

Exceeds target by 1725 ms

RAM

287 kB of 1536 kB

ROM

173 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 1:47:34 PM

96%

grayscale-mobilenetv1-9b3

PERFORMANCE

LATENCY

2345 ms of 100 ms

Exceeds target by 2245 ms

RAM

99 kB of 1536 kB

ROM

328 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 2:36:14 PM

95%

rgb-mobilenetv2-0de

PERFORMANCE

LATENCY

17551 ms of 100 ms

Exceeds target by 17451 ms

RAM

738 kB of 1536 kB

ROM

586 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

160 |

160

IMAGE

RGB

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 1:11:56 PM

95%

rgb-mobilenetv2-204

PERFORMANCE

LATENCY

1838 ms of 100 ms

Exceeds target by 1738 ms

RAM

287 kB of 1536 kB

ROM

173 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

RGB

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 2:24:41 PM

95%

grayscale-mobilenetv1-380

PERFORMANCE

LATENCY

2345 ms of 100 ms

Exceeds target by 2245 ms

RAM

135 kB of 1536 kB

ROM

312 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 12:55:28 PM

95%

rgb-mobilenetv2-946

PERFORMANCE

LATENCY

1838 ms of 100 ms

Exceeds target by 1738 ms

RAM

287 kB of 1536 kB

ROM

173 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

RGB

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 1:30:42 PM

95%

rgb-conv2d-325

PERFORMANCE

LATENCY

133 ms of 100 ms

Exceeds target by 33 ms

RAM

20 kB of 1536 kB

ROM

72 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

32 |

32

IMAGE

RGB

ACCURACY (KERAS)

12/15/2021, 2:29:17 PM

95%

grayscale-mobilenetv2-838

PERFORMANCE

LATENCY

6291 ms of 100 ms

Exceeds target by 6191 ms

RAM

351 kB of 1536 kB

ROM

585 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 3:39:04 PM

94%

grayscale-mobilenetv1-26b

PERFORMANCE

LATENCY

1472 ms of 100 ms

Exceeds target by 1372 ms

RAM

111 kB of 1536 kB

ROM

228 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 1:20:36 PM

94%

grayscale-mobilenetv1-a98

PERFORMANCE

LATENCY

1472 ms of 100 ms

Exceeds target by 1372 ms

RAM

111 kB of 1536 kB

ROM

228 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 1:29:51 PM

94%

rgb-mobilenetv2-9d5

PERFORMANCE

LATENCY

2143 ms of 100 ms

Exceeds target by 2043 ms

RAM

368 kB of 1536 kB

ROM

1208 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

RGB

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 2:35:00 PM

94%

grayscale-mobilenetv1-af9

PERFORMANCE

LATENCY

1472 ms of 100 ms

Exceeds target by 1372 ms

RAM

111 kB of 1536 kB

ROM

228 kB of 8096 kB

DSP NN Unused

IMAGE INPUT

96 |

96

IMAGE

Grayscale

ACCURACY (KERAS-TRANSFER-IMAGE)

12/15/2021, 12:49:28 PM

Training output

Hide

Click 'Run EON Tuner' to begin

EON Tuner settings

Find the optimal architecture for your machine learning model

The EON™ Tuner will evaluate many candidate model architectures (selected based on your target device and latency requirements) concurrently to help you find the best performing architecture for your application.

The search process can take up to 6 hours to complete. While the EON Tuner is running you can view the progress on this page at any time.

Name

Run #1:

Search space configuration

Configure the search space to narrow down your EON Tuner results. Include multiple parameter values to search for different combinations. To start, select a template below, a previous search space configuration, or add blocks manually.

Continuous motion

Classification

Detect cases where a motion is constantly playing.

Classification & anomaly

Detect cases where a motion is constantly playing

Motion events

Classification

Listen for short one-time occurances of motion

No previous runs available

INPUT

Time-series
Pre-processed features

DSP

Raw
Spectral analysis
IMU (Syntiant)

LEARN

Dense
Convolutional (1D)

ANOMALY

Anomaly (k-means)
Anomaly (GMM)

ORGANIZATION (DSP)

Add custom DSP block

ORGANIZATION (ML)

Add custom ML block

Do more with EON Tuner

Access more parallel EON Tuner jobs for faster results, a fully customizable search space, and more model variant options.

© 2024 EdgeImpulse Inc. All rights reserved

Show logs

Waiting...

Cancel job

Help & documentation Get help on the forum Send feedback Contact sales

Join us

Terms of Service Privacy policy

Send us your feedback!

We're always looking for ways to improve Edge Impulse. If you have any feedback, please let us know!

Subject

This field is required

How was your experience?

This field is required

Almost there!

You'll need a free Edge Impulse account to clone this project. Creating an account lets you add your own data, modify models, and join a community of thousands of embedded machine learning developers!

Sign up for free

I already have an account

Configure your target device and application budget

Target device

Define your target device requirements to inform model optimizations and performance calculations. No device yet? Use the default settings which you can change at any time.

Target device

Processor type

Processor family

Processor architecture

Clock rate

| MHz

Max

Accelerator

Device ID (optional)

Custom device name (optional)

Application budget

Specify the available RAM and ROM for the model's operation, along with the maximum allowed latency for your specific application. Not sure yet? Start with the defaults and modify them later on.

RAM

| KB

Max

ROM

| KB

Max

Latency

| ms

Max

Choose your pricing

YEARLY

$400 /month

billed annually

SAVE 15%

MONTHLY

$475 /month

billed monthly

Additional usage

Professional Plan includes 1,000 compute minutes per month. Additional usage will be charged at $0.10 per minute, billed monthly.

Upgrade to Enterprise

Why is this the right time for your team to invest in edge AI? What best describes your use case? What is your expected timeline for solving your problem?