Image Embedding

Generate image embeddings for large-scale datasets using NeMo Curator’s built-in and custom embedders. Image embeddings enable downstream tasks such as classification, filtering, duplicate removal, and similarity search.

How It Works

Image embedding in NeMo Curator typically follows these steps:

1. Load your dataset using ImageTextPairDataset

2. Select and configure an embedder (for example, TimmImageEmbedder)

3. Apply the embedder to generate embeddings for each image

4. Save the resulting dataset with embeddings for downstream use

You can use built-in embedders or implement your own for advanced use cases.

---

Available Embedding Tools

TimmImageEmbedder

Use state-of-the-art models from the PyTorch Image Models (timm) library for embedding generation

TimmImageEmbedder

Custom ImageEmbedder

Implement your own image embedding logic by subclassing the base class

Custom Image Embedders

---