Skip to content

Data Optimization

This chapter explains how to use the built-in Test tool of ATSVisionOps to optimize a dataset. This feature supports automated renaming, unified defect class names, and mapping modifications, all without damaging the original data.

test1.png

1. Default Test Logic (Direct Optimization)

If you do not need to make any changes to the current image names or defect names and only want to optimize the dataset, you can use the default operation.

  • Steps:
    1. Confirm that the input path is correct.
    2. Click the [Start Optimizing Dataset] button directly (no change options are checked by default).
  • Result:
    • After "Completed" is displayed, the system generates the image and JSON files in the corresponding Result Folder.
    • Data consistency: The images and JSON files correspond one-to-one and can be used directly for subsequent COCO format conversion or model training.

2. Batch Renaming Files

To make data management easier, you can rename the dataset globally, and the system will automatically update the reference fields inside the JSON files in sync.

  • Steps:
    1. Check the "Batch Rename Files" option.
    2. Enter the target prefix in the input box (e.g.: 111).
    3. Click [Start Optimizing Dataset].
  • Features:

    • Synchronized modification: It not only changes the file names on the physical disk (such as 111_000001.jpg), but also automatically corrects the imagePath field inside the JSON file, ensuring that no data associations are lost.

    • test3.png

    • test2.png

3. Defect Name Management

For cases where label names are non-standard or categories need to be merged, the system provides the following two modes:

A. Unify Defect Names (Global Overwrite)

Applicable when all defects of different categories in the current dataset need to be merged into a single new name.

  • Example: The original label is qx (defect), and you want to change it uniformly to bd (white spot).

  • Effect: After processing, the label field in every JSON file in the result folder will be changed to bd.

B. Map Specific Defects (Precise Replacement)

Applicable when only specific defect categories need to be changed while all other categories are kept unchanged.

  • Example: The dataset contains three labels, qx, zz, and bd, and only zz (impurity) needs to be changed to bszz (white impurity).

  • Operation: Check "Map Specific Defects", enter zz in the old name field and bszz in the new name field.

  • Effect: Only zz is replaced; all other labels remain unaffected.


Data safety: All optimization operations are performed in the "Result Folder" and will not modify your original input data.

test4.png