Data Cleansing

PCC Operator Overview

ModelArts data cleansing is implemented by the PCC operator. The dataset used for image classification or object detection may contain images that do not belong to the required categories. These images need to be removed to avoid interference to labeling and model training.

Description

**Table 1** Parameters of the PCC operator for data cleansing
Name	Mandatory	Default	Description
prototype_sample_path	Yes	None	Directory for storing positive data cleansing samples. The directory stores positive sample image files. The algorithm filters input data based on the positive sample images. That is, the data that is highly similar to the images in the prototype_sample_path directory is retained. Enter an existing OBS directory. The directory contains the provided positive sample images and starts with obs://, for example, obs://obs_bucket_name/folder_name.
criticism_sample_path	No	None	Directory for storing negative data cleansing samples. The directory stores negative sample image files. The algorithm filters input data based on the negative sample images. That is, the data that is less similar to the images in the criticism_sample_path directory is retained. It is recommended that this parameter be used together with prototype_sample_path to improve the accuracy of data cleansing. Enter an existing OBS directory that starts with obs://, for example, obs://obs_bucket_name/folder_name.
n_clusters	No	auto	Number of data sample types. The default value is auto. You can enter an integer less than the total number of samples or auto. auto indicates that the number of images in the positive sample directory is used as the number of data sample types.
simlarity_threshold	No	0.9	Similarity threshold. If the similarity between two images exceeds the threshold, the two images are regarded as similar. Otherwise, they are regarded as dissimilar. The value ranges from 0 to 1.
embedding_distance	No	0.2	Distance between sample features. If the feature distance between two images is smaller than the specified value, the two images are regarded as similar. Otherwise, the two images are regarded as dissimilar. The value ranges from 0 to 1.
do_validation	No	True	Indicates whether to validate data. The value can be True or False. True indicates that data is validated before cleansing. False indicates that data is cleansed only.

Operator Input Requirements

The following two types of operator input are available:

Datasets: Select a dataset and its version created on the ModelArts console from the drop-down list. Ensure that the dataset type be the same as the scenario type selected in this task.
OBSCatalog: Select either of the following storage structures:
- Only images: If the directory contains only images, the JPG, JPEG, PNG, and BMP formats are supported, and all images in the nested subdirectories are read.
- Images and labels: The structure varies depending on the scenario type.
  The following shows the directory structure in the image classification scenario. The following directory structure supports only single-label scenarios.
```
input_path/
    --label1/
        ----1.jpg
    --label2/
        ----2.jpg
    --../
```
  The following shows the directory structure in the object detection scenario. Images in JPG, JPEG, PNG, and BMP formats are supported. XML files are standard PACAL VOC files.
```
input_path/
    --1.jpg
    --1.xml
    --2.jpg
    --2.xml
    ...
```

Output Description

Image classification

The output directory structure is as follows:

output_path/
    --Data/
        ----class1/  # If the input data has labeling information, the information is also output. class1 indicates the labeling class.
            ------1.jpg
        ----class2/
            ------2.jpg
        ----3.jpg
    --output.manifest

A manifest file example is as follows:

{
	"id": "xss",
	"source": "obs://home/fc8e2688015d4a1784dcbda44d840307_14.jpg",
	"usage": "train", 
	"annotation": [
		{
			"name": "Cat", 
			"type": "modelarts/image_classification"
		}
	]
}

Object detection

The output directory structure is as follows:

output_path/
    --Data/
        ----1.jpg
        ----1.xml  # If the input data has labeling information, the information is also output. xml indicates the label file.
        ----2.jpg
        ----3.jpg
    --output.manifest

A manifest file example is as follows:

{
	"source":"obs://fake/be462ea9c5abc09f.jpg",
	"annotation":[
		{
			"annotation-loc":"obs://fake/be462ea9c5abc09f.xml",
			"type":"modelarts/object_detection",
			"annotation-format":"PASCAL VOC",
			"annotated-by":"modelarts/hard_example_algo"
		}
	]
}

Parent topic: Description of Built-in Operators for Data Processing

Previous topic: Data Validation

Next topic: Data Selection