The "Extract Subset" page within Dataset Tools empowers you to create smaller, focused datasets from larger ones by applying advanced filtering based on specific tags, words, or terms. This feature allows you to curate datasets that meet your project's precise requirements. This is specially useful when fine tuning large datasets; but somehow your model is still struggling with specific concepts... so you can easily extract a part of your dataset for that specific concept and train only that small portion of the dataset.

The "Extract Subset" feature provides the following functionalities:

1. Advanced Search Configuration: Utilize a buildable search configuration to filter images based on specific tags, words, or terms:

   • Search for a specific tag using "tag."
   • Exclude a specified tag using "!tag."
   • Use the "AND" operator (in uppercase) to find images containing one or more tags at the same time. For example, "tag1 AND tag2.".
   • Example Search Configuration: "tag1, tag2, !tag3, tag4 AND tag5."

2. Selective Search Scope: Choose to search within .caption files, .txt files, or both, allowing you to narrow your focus to specific metadata sources.

3. Exact or Partial Match: Optionally perform searches for exact matches or partial matches; enhancing flexibility in your filtering criteria.

1. Select Input and Output Folders: Specify the input folder containing your images and the output folder where a copy of the extracted subset will be saved (the original dataset wont be modified).

2. Configure Search Filters: Use the advanced search configuration to define your filtering criteria based on tags, words, or terms.

3. Specify Search Scope: Decide whether you want to search within .caption files, .txt files, or both.

4. Choose Match Type: Select whether you want to perform exact matches, partial matches, or a combination of both for your filtering criteria.

5. Extract Subset: Click the "Filter" button to initiate the extraction process. The tool will create a new dataset based on your specified filters.

The "Extract Subset" feature is a powerful tool for creating smaller, focused datasets from larger ones, enabling you to efficiently curate datasets that align with your project's objectives.

• Experiment with different search configurations to extract subsets tailored to your specific needs.
• Use the exact match option for precise filtering, and the partial match option for broader inclusion.