Data Augmentation Tool for Tag Wrinkles

The path of the data augmentation tool is ./dataset_tools/tag_wrinkle_augmentation/tag_paste_data_augmentation.py. The script supports the input of the tag image folder path, parameter configurations, output of the augmented images, and generation of images with wrinkles and tilts based on the configured parameters.

To generate a simulation image, you need to manually configure parameters. You can modify the parameters to change the authenticity of the generated wrinkles and bubbles. You are advised to use the default parameters. For details, see the following table.

**Table 1** Recommended parameter settings
Parameter	Recommended Wrinkle Feature Value	Recommended Bubble Feature Value
area_ratio	[0.0001,0.01]	[0.005,0.02]
aspect_ratio	[0.025,0.2]	[0.33,3.0]
rotation	[-90,90]	[-90,90]
up_ratio	[1.5,3.5]	[1.1,1.3]
down_ratio	[0.5,0.8]	[0.5,0.8]
area_ratio_thres	0.01	0.02
same_ratio_enable	False	True

Example of default parameters for generating wrinkles:

python3 ./dataset_tools/tag_wrinkle_augmentation/tag_paste_data_augmentation.py --input_dataset_path=/data/small --output_dataset_path=/data/output

Example of manually modified parameters:

python3 ./dataset_tools/tag_wrinkle_augmentation/tag_paste_data_augmentation.py --input_dataset_path=/img/path --output_dataset_path=/data/output --area_ratio=[0.0001,0.01] --aspect_ratio=[0.025,0.2] --rotation=[-90,90] --up_ratio=[1.5,3.5] --down_ratio=[0.5,0.8] --area_ratio_thres=0.01 --same_ratio_enable=false --each_type_generate_nums=1

A directory is created in the generation folder to store the image data generated based on user-defined parameters. Table 2 describes the parameters.

If there are multiple types of tags and features, you need to manually configure script parameters to generate tags and features separately.

**Table 2** Execution parameters of the dataset augmentation tool
Parameter	Type	Value Range	Default Value	Description
--input_dataset_path	String	-	Mandatory	Input path of the tag image folder.
--output_dataset_path	String	-	Mandatory	Output path of the image folder, which cannot be the same as the input path. The script creates images and annotations folders in output_dataset_path, and saves the generated image data to images and the labeled data to the annotations folder in ImageNet format.
--each_type_generate_nums	Integer	> 0	1	Setting of the quantity of images with wrinkles or bubbles.
--area_ratio	List	[0.0001, 0.02]	[0.0001,0.01]	Area ratio range of a single defect.
--aspect_ratio	List	[0.005, 3.0]	[0.025, 0.2]	Aspect ratio range of a single defect.
--rotation	List	[-90, 90]	[-90, 90]	Rotation ratio range of a single defect.
--up_ratio	List	[1.0, 3.5]	[1.5, 3.5]	Luminance transformation ratio range of the upper part of the image.
--down_ratio	List	[0.5, 1.0]	[0.5, 0.8]	Luminance transformation ratio range of the lower part of the image.
--area_ratio_thres	Float	Greater than 0 and less than 1	0.01	Threshold of the total defect area ratio. The recommended value range is [0.005,0.02].
--same_ratio_enable	Bool	true, false	false	The lower part of the image synchronizes with the upper part.
--timeout	Integer	-	300	The default timeout period is 5 minutes.

Parent topic: Dataset Creation for Tag Defect Detection and Classification