cryptolab-playground/amazon-products-clip-vit-b-32

Amazon Products CLIP ViT-B/32

Dataset Description

This dataset contains pre-computed embeddings of product image and text pairs from Amazon, designed for evaluating vector database performance on multi-modal datasets. The embeddings are generated using OpenAI's CLIP-ViT-B/32 model.

Purpose

Benchmark dataset for evaluating vector database performance, specifically designed for use with VectorDBBench.

Dataset Summary

• Total Training Samples: 400,000 images
• Test Queries: 1,000 texts
• Ground Truth: Top-1000 nearest neighbors per query
• Embedding Dimension: 512
• Embedding Model: CLIP-ViT-B-32
• Source Data: Amazon-Products-Eval

Dataset Structure

Data Splits

Split	Samples	Description
train	400,000	Training image embeddings (random sample from source image)
test	1,000	Test query embeddings (random sample from corresponding source text)
neighbors.parquet	1,000	Top-1000 nearest neighbors for each test query

Data Fields

train & test

• id (int64): Unique identifier for each product
• emb (List[float64]): 512-dimensional L2-normalized embedding vector

neighbors.parquet

• id (int64): Query identifier (matches test)
• neighbors_id (List[int64]): List of 1,000 nearest neighbor IDs from train set

Dataset Creation

Source Data

The dataset is derived from approximately 1M products image-text pairs from Marqo/amazon-products-eval:

• Train: 400,000 randomly sampled images
• Test: 1,000 randomly sampled from corresponding texts

Preprocessing

1. Data Preparation: images and text embeddings were generated by embedding model
2. Normalization: All embeddings are L2-normalized

Embedding Generation

• Model: sentence-transformers/clip-ViT-B-32
• Dimension: 512
• Normalization: L2-normalized

Ground Truth Generation

Ground truth nearest neighbors were computed using:

• Method: Flat search (brute-force)
• Metric: Cosine similarity
• K: Top-1000 neighbors per query

Usage

Loading the Dataset

from datasets import load_dataset
import pandas as pd

# Load train and test splits
dataset = load_dataset("cryptolab-playground/amazon-products-clip-vit-b-32")

Evaluation Example

import numpy as np
from datasets import load_dataset
import pandas as pd

# Load data
dataset = load_dataset("cryptolab-playground/amazon-products-clip-vit-b-32")

neighbors = pd.read_parquet(
    "hf://datasets/cryptolab-playground/amazon-products-clip-vit-b-32/neighbors.parquet"
)

# Convert to numpy arrays
train_embeddings = np.array(dataset['train']['emb'])
test_embeddings = np.array(dataset['test']['emb'])

# Example: Compute recall@10
def compute_recall_at_k(retrieved_ids, neighbors_ids, k=10):
    """
    Compute Recall@K
    
    Args:
        retrieved_ids: List of retrieved neighbor IDs
        neighbors_ids: List of ground truth neighbor IDs
        k: Number of top results to consider
    """
    retrieved_k = set(retrieved_ids[:k])
    neighbors_k = set(neighbors_ids[:k])
    
    if len(neighbors_k) == 0:
        return 0.0
    
    return len(retrieved_k & neighbors_k) / len(neighbors_k)

# Use with your vector database
# ... insert your vector DB search code here ...

Use Cases

• Vector database performance benchmarking
• Approximate nearest neighbor (ANN) algorithm evaluation
• Retrieval system testing on product images

Limitations

• Domain-Specific: Optimized for product images; may not generalize to other domains
• Language: English only
• Ground Truth: Based on cosine similarity with embeddings, not human relevance judgments

License

Apache 2.0 (same as source dataset)

Citation

If you use this dataset, please cite:

@dataset{cryptolab-playground/amazon-products-clip-vit-b-32,
  author = {CryptoLab, Inc.},
  title = {Amazon Products CLIP ViT-B/32},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/cryptolab-playground/amazon-products-clip-vit-b-32}
}

Source Dataset Citation

@software{zhu2024marqoecommembed_2024,
    author = {Tianyu Zhu and and Jesse Clark},
    month = oct,
    title = {{Marqo Ecommerce Embeddings - Foundation Model for Product Embeddings}},
    url = {https://github.com/marqo-ai/marqo-ecommerce-embeddings/},
    version = {1.0.0},
    year = {2024}
}

Embedding Model Citation

@misc{clipvitb32,
  title={CLIP ViT-B/32},
  author={Open AI},
  year={2021},
  url={https://huggingface.co/sentence-transformers/clip-ViT-B-32}
}

Acknowledgments

• Original dataset: Marqo/amazon-products-eval
• Embedding model: sentence-transformers/clip-ViT-B-32
• Benchmark framework: VectorDBBench