To harness the powerful capabilities of Visual Search in HawkSearch, a structured process needs to be followed. This process ensures that the feature is not only enabled correctly but also configured to meet your specific search needs.

Here’s a step-by-step instruction:

Feature Activation

The Visual Search feature must first be enabled by our internal personnel. To initiate this process, please contact the our CSD team through email or fill out the interest form within the workbench. They will work with you to activate the feature on your behalf, ensuring a smooth and efficient setup.

Configure Fields for Visual Search

Once the Visual Search feature is enabled, the next step is to set up the specific field for visual search functionality. This setup involves identifying and selecting the field within your index that stores the image URLs, which are crucial for creating search embeddings. Please be aware that the system currently allows only one field for visual search. It is essential to select the most suitable field containing the image URLs to ensure optimal performance of the Visual Search feature. For comprehensive guidance and best practice on completing the configuration, please refer to this detailed article.

Perform a Full Index Rebuild

After configuring the fields, a crucial step is to rebuild the full index. This process allows the system to create the necessary embeddings for the configured field, ensuring that the Visual Search can function as intended. Rebuilding the full index integrates all the setup changes and prepares the system for executing visual searches. Please note that this process involves creating a new index and updating all related components, as detailed in the provided documentation.

Making Visual Search Requests

With the feature enabled, fields configured, and full index rebuilt, you’re now ready to utilize the Visual Search! There are two ways to perform a visual search:

Visual Search: For this method, you will need to upload an image to serve as your search query. It's important to note that the image must be encoded to base64 format before sending the request through the "ImageData" field. The system then analyzes this image to identify and return items that are visually similar to the content depicted in the image.

Example Request:
```
{
    "ClientGuid": "xxxxx",
    "RequestType":"ImageSearch",
    "kValue":3,
    "ImageData": "data:image/png;base64,<<base64 data here>>"
}
```

Image Search: Alternatively, you can describe an image in text within the "Keyword" field when the "RequestType" is set to "ImageSearch". This option is useful when you do not have an image but can provide a detailed description of what you’re searching for. The system interprets your description to find matching or relevant items.

Example Request:
```
{
    "ClientGuid": "xxxxx",
    "RequestType":"ImageSearch",
    "kValue":3,
    "Keyword": "Show me the picture with a woman climbing a cliff"
}
```

By following these steps, you can fully enable and utilize the Visual Search, significantly enhancing the search experience on your platform by allowing for more nuanced and visually oriented search queries. For more technical details regarding the visual search request, please refer to this article or check out the API reference.

Updates in ImageText in Athena release (June 2024)

Integrating ImageText Field into Keyword Field for Image Finder

This update refines the Image Finder's search functionality by replacing the ImageText field with the Keyword field in the Search API call. This change enhances the relevance and targeting of visual search results by allowing the Keyword field to complement the RequestType, thereby accommodating existing Hawk features such as merchandising rules. The integration aims to provide more accurate and contextually relevant images.

Backward Compatibility

Mapping ImageText to Keyword:
1. When only the ImageText field is provided in the request:
  1. Use the ImageText value for the search by internally assigning it to the Keyword field.
2. This assignment allows existing searches to function as intended without any intervention.
Prioritizing Keyword over ImageText:
1. If both Keyword and ImageText fields are provided:
  1. Ignore ImageText entirely and process only the Keyword field for search purposes.

Keyword Provided	ImageText Provided	Action Performed
Yes	No	Use the value in `Keyword` for search.
No	Yes	Assign the `ImageText` value to the `Keyword` field.
Yes	Yes	Use `Keyword` for search, ignore `ImageText`
No	No	Either there’s no description for Image Finder request or the request is for Visual Search