Edge Impulse Inc. / Visual GMM cracks Public

Anomaly detection settings

Training settings

Training processor

Please provide a valid training processor option

Capacity

Backbone hyperparameters

Number of layers

Advanced training settings

Validation set size

Please provide a valid number for the train/validate split (between 0 and 1)

Split train/validation set on metadata key

Batch size

Auto-weight classes

Profile int8 model

Neural network architecture

Save

Input layer (25,600 features)

MobileNetV2 0.35

Gaussian Mixture Model (GMM)

Output layer (Anomaly score)

Model

Model version:

By definition there should be as few as possible anomalies in the training dataset, and thus accuracy is not calculated during training. Run Model testing to learn more about the model performance.

On-device performance

Engine:

Inferencing time

260 ms.

Peak RAM usage

2.4M

Flash usage

86.7K

Did you know? You can customize your model through the Expert view (click on to switch),
or can even bring your own model (in PyTorch, Keras or scikit-learn).

Model	Author
Gaussian Mixture Model (GMM) Officially supported Visual anomaly detection for low-dimensional images. Works best with images below 160x160 pixels in size or when defects are above 16x16 pixels in size. Lower memory footprint. Supports both RGB and grayscale.	Edge Impulse
Patchcore Officially supported Visual anomaly detection for high-dimensional or complex images, while also performing well with smaller images. Works better than GMM for images above 160x160 pixels in size or when defects are below 16x16 pixels in size. Higher memory footprint. Supports both RGB and grayscale.	Edge Impulse

Gaussian Mixture Model (GMM)
Officially supported

Visual anomaly detection for low-dimensional images. Works best with images below 160x160 pixels in size or when defects are above 16x16 pixels in size. Lower memory footprint. Supports both RGB and grayscale.

Patchcore
Officially supported

Visual anomaly detection for high-dimensional or complex images, while also performing well with smaller images. Works better than GMM for images above 160x160 pixels in size or when defects are below 16x16 pixels in size. Higher memory footprint. Supports both RGB and grayscale.

The backbone is the foundational element of a neural network model, responsible for extracting meaningful features from input data, enabling subsequent parts to perform anomaly scoring or classification.

Description	Author
EfficientNet V2B0 Uses around 755K RAM based on your input size, and between 240-1675K ROM depending on the number of layers, with default compiler settings. Supports RGB.	Edge Impulse
MobileNetV2 0.35 Uses around 636K RAM based on your input size, and between 69-134K ROM depending on the number of layers, with default compiler settings. Supports both RGB and grayscale.	Edge Impulse
MobileNetV2 0.50 Uses around 643K RAM based on your input size, and between 78-168K ROM depending on the number of layers, with default compiler settings. Supports both RGB and grayscale.	Edge Impulse
MobileNetV2 0.75 Uses around 1263K RAM based on your input size, and between 109-266K ROM depending on the number of layers, with default compiler settings. Supports both RGB and grayscale.	Edge Impulse
MobileNetV2 1.0 Uses around 1273K RAM based on your input size, and between 126-372K ROM depending on the number of layers, with default compiler settings. Supports both RGB and grayscale.	Edge Impulse
MobileNetV2 0.1 Uses around 622K RAM based on your input size, and 58K ROM, with default compiler settings. Supports both RGB and grayscale.	Edge Impulse

Uses around 755K RAM based on your input size, and between 240-1675K ROM depending on the number of layers, with default compiler settings. Supports RGB.

Uses around 636K RAM based on your input size, and between 69-134K ROM depending on the number of layers, with default compiler settings. Supports both RGB and grayscale.

Uses around 643K RAM based on your input size, and between 78-168K ROM depending on the number of layers, with default compiler settings. Supports both RGB and grayscale.

Uses around 1263K RAM based on your input size, and between 109-266K ROM depending on the number of layers, with default compiler settings. Supports both RGB and grayscale.

Uses around 1273K RAM based on your input size, and between 126-372K ROM depending on the number of layers, with default compiler settings. Supports both RGB and grayscale.

Uses around 622K RAM based on your input size, and 58K ROM, with default compiler settings. Supports both RGB and grayscale.

Some backbones have been disabled and hidden because they are not supported by the selected scoring function. Show all backbones anyway

The scoring function operates on the features extracted by the backbone, and returns anomaly scores for each segment of the image.

Description	Author	Recommended
Gaussian Mixture Model (GMM) Visual anomaly detection for low-dimensional images. Works best with images below 160x160 pixels in size or when defects are above 16x16 pixels in size. Lower memory footprint. Supports both RGB and grayscale.	Edge Impulse
Patchcore Visual anomaly detection for high-dimensional or complex images, while also performing well with smaller images. Works better than GMM for images above 160x160 pixels in size or when defects are below 16x16 pixels in size. Higher memory footprint. Supports both RGB and grayscale.	Edge Impulse

Gaussian Mixture Model (GMM)

These users will be notified over email when this job finishes.

	Name
	Daniel Situnayake Edge Impulse staff
	Aurelien Lequertier Edge Impulse staff
	Jenny Plunkett Edge Impulse staff
	Louis Moreau Edge Impulse staff
	Alessandro Edge Impulse staff
	Mateusz Majchrzycki Edge Impulse staff
	D Dmitry Edge Impulse staff
	Sara Edge Impulse staff
	Ivan Turasov Edge Impulse staff
	ML Marco Levrero Edge Impulse staff

Configure your target device for model performance calculations, and to enable target specific optimizations.

Every learning block has a threshold. This can be the minimum confidence that a neural network needs to have, or the maximum anomaly score before a sample is tagged as an anomaly. You can configure these thresholds to tweak the sensitivity of these learning blocks. This affects both live classification and model testing.

Visual anomaly detection

Minimum score before tagging as anomaly

Suggested value is 16.6, based on the top anomaly scores in the training dataset.

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

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!

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 family

Clock rate

| MHz

Max

Custom device name

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

YEARLY

$400 /month

billed annually

SAVE 15%

MONTHLY

$475 /month

billed monthly

Anomaly detection settings

Training settings

Backbone hyperparameters

Augmentation settings

Advanced training settings

Neural network architecture

Model

Confusion matrix (validation set)

Data explorer (full training set)

On-device performance

Inferencing time

Peak RAM usage

Flash usage

Confusion matrix (validation set)

Data explorer (full training set)

On-device performance

Inferencing time

Peak RAM usage

Flash usage

Gaussian Mixture Model (GMM)
Officially supported

Patchcore
Officially supported

EfficientNet V2B0

MobileNetV2 0.35

MobileNetV2 0.50

MobileNetV2 0.75

MobileNetV2 1.0

MobileNetV2 0.1

Gaussian Mixture Model (GMM)

Patchcore

Anomaly detection settings

Training settings

Backbone hyperparameters

Augmentation settings

Advanced training settings

Neural network architecture

Model

Confusion matrix (validation set)

Data explorer (full training set)

Settings

On-device performance

Inferencing time

Peak RAM usage

Flash usage

Confusion matrix (validation set)

Data explorer (full training set)

Settings

On-device performance

Inferencing time

Peak RAM usage

Flash usage