Index Custom Embeddings

​

Custom Embeddings Support

Encord enables the use of custom embeddings for images, image sequences, image groups, and individual video frames. Custom embeddings allow you to incorporate your own feature representations into Encord’s platform, enhancing capabilities like similarity search, visualizations, and data filtering. This flexibility supports more advanced workflows and deeper insights tailored to your specific use cases.

​

Use Custom Embeddings in Index

To bring your custom embeddings into Encord, you first need to create a key in your metadata schema. After the key is in your schema, you can import your custom embeddings.

To use custom embeddings in Index:

1. Create a new embedding type in your Schema.
2. Upload your embeddings.
3. Select your custom embeddings from the Embeddings view.

​

Step 1: Create a New Embedding Type

A key is required in your custom metadata schema for your embeddings. You can use any string as the key for your embeddings. We strongly recommend that you use a string that is meaningful.

If you do not include a key in your metadata schema, your imported embeddings are treated as strings.

Use add_embedding to add an embedding to your metadata schema.

​

Step 2: Upload Embeddings

With the key in the custom metadata schema ready, we can now import our embeddings.

Active embeddings MUST be of dimension 512. Index supports custom embeddings from a range of 1 to 4096.

You can import embeddings after you have imported your data or during your data import.

If config is not specified, the sampling_rate is 1 frame per second, and the keyframe_mode is frame.

# Import dependencies
from encord import EncordUserClient
from encord.http.bundle import Bundle

# Authentication
SSH_PATH = "<file-path-to-ssh-private-key>"

# Authenticate with Encord using the path to your private key
user_client: EncordUserClient = EncordUserClient.create_with_ssh_private_key(
    ssh_private_key_path=SSH_PATH,
)

# Define a dictionary with item UUIDs and their respective metadata updates
updates = {
    "<data-ID-1>": {"<my-embedding>": [1.0, 2.0, 3.0]},
    "<data-ID-2>": {"<my-embedding>": [1.0, 2.0, 3.0]}
}

# Use the Bundle context manager
with Bundle() as bundle:
    # Update the storage items based on the dictionary
    for item_uuid, metadata_update in updates.items():
        item = user_client.get_storage_item(item_uuid=item_uuid)

        # Make a copy of the current metadata and update it with the new metadata
        curr_metadata = item.client_metadata.copy()
        curr_metadata.update(metadata_update)

        # Update the item with the new metadata and bundle
        item.update(client_metadata=curr_metadata, bundle=bundle)

"config": {
    "sampling_rate": "<samples-per-second>",
    "keyframe_mode": "frame" or "seconds",
},

​

Step 3: Select your Custom Embeddings

After you import your custom embeddings they are available for:

• Filtering using custom embeddings

• Similarity searches using your custom embeddings

• Embedding view and 2D plots with selection based workflows

​

Upgrade your Top-Level Folder

Before you can perform filtering, use similarity searches, or use embedding plots with your custom embeddings, you need to configure your top-level Folder in Files.

To configure Folders for Embedding Plots:

1. Go to Index > Files.

   A list of Folders available to you appears on the My Files page.

2. Do one of the following:

   • Select the check box for the Folder.
   • Click into the Folder.

   The Upgrade Folder button appears.

3. Click Upgrade Folder.

   The Folder upgrades dialog appears.

4. Expand the Custom Embeddings drop down.

5. Select a custom embedding from the list.

6. Click Add.

   The custom embedding appears under Custom Embeddings.

   You can add multiple embeddings. Only one embedding can be active in Index at a time.

7. Expand your selected custom embedding.

8. Select any of the following:

   • Similarity search
   • Compute UMAP Embedding Reduction
   • Compute Advanced Quality Metrics

9. Click Save and process changes.

   A dialog appears informing you that the folder upgrade was successful. You are now ready to use your custom embeddings.

​

Filtering with Custom Embeddings

Upgrade your top-level folder before trying to filter.

1. Click the Filters dropdown or press F. The Filters menu appears.

2. Click Custom Embeddings from the menu.

3. Select your custom embedding to filter your data.

4. Select True to display images, frames, or videos with the custom embeddings.

​

Similarity Searches with Custom Embeddings

Upgrade your top-level folder before trying to perform a similarity search.

1. Click the Embeddings icon in the Explorer.

   The Embeddings screen appears.

2. Select the embedding you want to use from the Select custom embeddings menu.

3. Click the Grid icon.

4. Hover over an image or frame with the custom embedding.

5. Click the Similarity Search icon. Images and Frames sort according to similarity.

​

Adjust Similarity Search Distance

1. Click the Embeddings icon in the Explorer. The Embeddings screen appears.

2. Select the embedding you want to use from the Select custom embeddings menu.

3. Click the Grid icon.

4. Hover over an image or frame with the custom embedding.

5. Click the Similarity Search icon. Images and Frames sort according to similarity AND a Distance filter appears.

6. Adjust the Distance filter slider to change the similarity search results.

​

Embedding View with Index

Upgrade your top-level folder before trying to view embedding plots.

Encord Index incorporates embedding plots — a two-dimensional visualization technique employed to represent intricate, high-dimensional data in a more comprehensible and visually coherent manner. This technique reduces data dimensionality while preserving the inherent structure and patterns within the original data.

The embedding plot aids in identifying interesting/noteworthy clusters, inspecting outliers, and excluding unwanted samples.

Use Custom Embedding Plots

Notice how images are clustered around certain regions. By defining a rectangular area on the plot, users can quickly isolate and analyze data points within that defined region. This approach facilitates the exploration of commonalities among these samples.

Hover over clusters or individual data points on the plot to visually check frames.

Upon selecting a region, the content within the Explorer page adjusts accordingly. Various actions can be executed with the chosen group:

• Use Collections to tag and group images.
• Establish subsets similar to these and then conduct comparisons.