Generate Annotated Dataset

Feature Description

This operator is used to convert annotation data of a specific format into COCO dataset format, which can be directly uploaded to the CVAT annotation platform for training deep learning models.

Usage Scenarios

Preparing deep learning data, which can be further used to optimize annotations and manage model data.

Inputs and Outputs

Input Items

Save Path: Specifies the folder where the original annotation data is located. The operator will look for .png files in this folder and expects each .png file to have a corresponding .json file with the same name (e.g., image1.png corresponds to image1.json).

If this input port is connected to data (i.e., a path is provided), this path will override the "Save Path" parameter set in the parameter panel as the read path for the original data.

If this input port is not connected to data (is empty), the operator will use the path specified by "Save Path" in the parameter panel as the read path for the original data.

Output Items

File Name: The main name of the actually saved COCO dataset file.

Parameter Description

The Annotation name parameter must be provided as an array in the correct format. Each custom category name must be enclosed in English double quotes ("), for example: ["CategoryA", "CategoryB"]. Otherwise, a "Formatting failed" prompt will be displayed.

Save Path

Parameter Description

Used to specify the target folder path for saving files.

Adjustment Notes

  • Specify Path: Directly enter a valid folder path, or click the icon on the right to select a folder path.

  • Do Not Specify Path: Leave it blank. The operator will, by default, create a subfolder named file_save_dir in the .transfertech atom folder as the storage path. This is convenient for temporary saving or when the specific storage location is not critical.

  • Path Check: The operator will check if the specified path is valid. If the path points to an existing file (not a folder), or if the path is invalid, the operator will report an error. If the specified path does not exist, it will attempt to create it.

Save Name

Parameter Description

Specifies the save file name; no need to include the file extension.

Adjustment Notes

It is recommended to use a descriptive name for easy retrieval and management later. Avoid using special characters.

Default name: data

Save Only Single File

Parameter Description

Sets the file saving strategy, either saving as a single file or multiple files.

Adjustment Notes

  • Disabled (default): Generates multiple files, automatically adding a timestamp after the specified file name to avoid duplication. Suitable for scenarios requiring historical data or multiple run results.

  • Enabled: The operator will attempt to use the specified name as the file name to save data each time it runs. If the file already exists, the new data will overwrite the old file. Suitable for scenarios requiring a fixed file name output, e.g., for overwriting the previous result.

Maximum Storable Memory

Parameter Description

Sets the maximum total size of files allowed in the save directory. When the total size of relevant files in the folder exceeds this limit, the software will attempt to delete the oldest files of the same type in that directory to free up space.

Adjustment Notes

Please set this value based on available disk space and data retention needs:

  • Smaller value: Will delete old files more frequently, saving disk space, but may lead to faster loss of historical data.

  • Larger value: Allows more historical data to be saved but will consume more disk space.

Parameter Range

[0.00001, 10], Default: 10, Unit: GB

Annotation Name

Parameter Description

A list of strings used to define all possible object class names in the dataset. Each string in the list represents a class.

Adjustment Notes

  • Custom Classes: Customize this list by setting class names according to your needs. For example, if your project detects "cat", "dog", "person", you can set it to ["cat", "dog", "person"].

  • Class ID Mapping: The operator will automatically map the first name in the list to ID 1, the second to ID 2, and so on. However, in the annotation file, the label field will correspond to the class name using a zero-based numeric string.

  • Default Value: If this list is empty, the operator will default to using ["0", "1", "2", "3", "4"] as class names, supporting up to 5 classes.