Indad

Overviews

INDAD is an Industrial KNN-based Anomaly Detection.

Original git: https://github.com/rvorias/ind_knn_ad.git

Example Detect Anomaly using PatchCore

The packages contain 3 different algorithms:

1. SPADE (2021)
2. PaDiM (2020)
3. PatchCore (2021): Recommended

In this document, we will show you how to use HACHIX version of these algorithms.

APIs

Model

Bases: BaseModel

This is the base model for INDAD

Parameters:

Name	Type	Description	Default
BaseModel	Class	Common class for all ecos_core models	required

Returns: None

__init__(opt)

This function is called when creating a new instance of INDAD Model

Parameters:

Name	Type	Description	Default
opt	NameSpace	The default options neccessary for initializing the model. opt should includes these attributes: class_names, weight, device.	required

build_model(custom_weight=None)

Build model in ensemble model. Type of model is defined in opt.json

Parameters:

Name	Type	Description	Default
custom_weight(str)		weight to load that not defined in opt. Default: None.	required

Returns:

Type	Description
Model	Model after built

Examples:

>>> from argparse import ArgumentParser
>>> from ecos_core.indad.indad import Model
>>> parser = ArgumentParser()
>>> opts = parser.parse_args()
>>> with open(opt_path, 'r') as f:
>>>     opts.__dict__ = json.load(f)
>>> model = Model(opts)
>>> #Build model with default options
>>> model.build_model()
>>> #Build model with custom weight
>>> model.build_model(custom_weight=<PATH_TO_WEIGHT>)

calculate_threshold_score(weight_path, valid_path)

calculate the threshold score from the validation dataset

Parameters:

Name	Type	Description	Default
weight_path	str	weight path	required
valid_path	str	validation images folder path	required

Returns:

Name	Type	Description
List		highest threshold score and pixel score
List		score per image

create_augment_images(good_folder_path, image_files, number_of_copy) staticmethod

Augment the images in a directory randomly, then generate training dataset

Parameters:

Name	Type	Description	Default
good_folder_path	str	The path to the directory containing the images	required
image_files	list	list of image files	required
number_of_copy	int	maximum number of copies per image to be augmented	required

create_validation_folder(valid_folder_path) staticmethod

create validation folder if it doesn't exist or not qualified

Args: valid_folder_path (str): description

Returns:

Name	Type	Description
Bool		True: Exist and qualified False: Not exist or not qualified

generate_instance_model(method) staticmethod

generate instance model

Parameters:

Name	Type	Description	Default
method	str	spade or padim or patchcore or amazon_patchcore method	required

Returns:

Name	Type	Description
object		instance of method

generate_training_data(train_path, number_of_copy)

Augment the images in a directory randomly, then generate training dataset

Parameters:

Name	Type	Description	Default
train_path	str	The path to the directory containing the images to be augmented	required
number_of_copy	int	maximum number of copies per image to be augmented	required

generate_validation_dataset(base_path, valid_folder_path, split_percentage) staticmethod

Create a validation dataset based on training dataset

Parameters:

Name	Type	Description	Default
base_path	str	path to data directory	required
valid_folder_path	str	validation folder path	required
split_percentage	float	ratio of validation dataset / training dataset	required

get_transform()

Transform input by ImageNet normalization.

Returns:

Name	Type	Description
Function		A function for transform a single image from path

load_dataset(good_folder_path) staticmethod

load data from specified folder, return list of image files name and number of images

Parameters:

Name	Type	Description	Default
good_folder_path	str	The path to the directory containing the images	required

Returns:

Name	Type	Description
number		number of images
list		list of image files

load_opt(opt_path) staticmethod

load opt file and return data

Parameters:

Name	Type	Description	Default
opt_path	str	opt.json path	required

Raises:

Type	Description
FileNotFoundError	description

Returns:

Name	Type	Description
dict		opt configuration

load_threshold_score(weight_path, valid_path)

read threshold_score file and return threshold score calculated from validation dataset if opt.json is exist
if not exist it will calculate threshold score from validation dataset

Parameters:

Name	Type	Description	Default
weight_path	str	weight path	required
valid_path	str	validation images folder path	required

Returns:

Name	Type	Description
List		threshold score and pixel score

process_predictions(pred, raw_image_path, raw_image, image_transform, save_image_path)

After processing for prediction

Parameters:

Name	Type	Description	Default
pred	_type_	description	required
image_transform	Array	Original image after transformation	required
raw_image	Image	PIL image	required
save_image_path	str	Path to save image	required

Returns:

Type	Description
str	Path to the saved image

save_threshold_score(threshold, threshold_pixel_score=0.0)

save to opt.json

Parameters:

Name	Type	Description	Default
threshold	int	threshold score	required
threshold_pixel_score(float)		threshold for pixel level score. Default 0.0	required