README.md

---
license: cc-by-4.0
configs:
- config_name: embeddings
  data_files: data/*.parquet
- config_name: id_mapping
  data_files: id_mapping/*.parquet
task_categories:
- image-to-text
- image-to-image
tags:
- images
- CLIP
- embeddings
- FAISS
size_categories:
- 1M<n<10M
---

<p align="center">
    <a href="https://github.com/ml6team/fondant">
        <img src="https://raw.githubusercontent.com/ml6team/fondant/main/docs/art/fondant_banner.svg" width="600px"/>
    </a>
</p>

<p align="center">
    <i>
        <b>Production-ready</b> 
        data processing made 
        <b>easy</b> 
        and 
        <b>shareable</b>
    </i>
    <br>
    <a href="http://fondant.ai"><strong>Explore the Fondant docs »</strong></a>
    <a href="https://discord.gg/HnTdWhydGp"><img alt="Discord" src="https://dcbadge.vercel.app/api/server/HnTdWhydGp?style=flat-square"></a>
</p>

# Dataset Card for fondant-ai/datacomp-small-clip

<!-- Provide a quick summary of the dataset. -->

This is a dataset containing image urls and their CLIP embeddings, based on the [datacomp_small](https://huggingface.co/datasets/mlfoundations/datacomp_small) dataset, and processed with [fondant](https://github.com/ml6team/fondant).

## Dataset Details

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

Large (image) datasets are often unwieldy to use due to their sheer size. Assume for instance
that we would like to extract all the cat images from such a dataset. We would have to look at
every image to classify if it's a cat image or not. And if we want to extract all the dog images
next, we again need to look at every image.

Instead, we can look at every image once, and calculate a (CLIP) embedding representing its
content. Combining these embeddings into an index, we can efficiently search through the dataset
with a query, finding specific images, without having to look at each one.

![CLIP index](https://cdn-uploads.huggingface.co/production/uploads/6454cb0e1a543cf97b1b6fd6/Mgl9UAqiwJrV4WDb8Y2-k.png)

This is what LAION did for their [LAION-5b dataset](https://laion.ai/blog/laion-5b/), which made 
it possible to use, like we did in our 
[ControlNet example](https://github.com/ml6team/fondant-usecase-controlnet). 
Unfortunately, the LAION-5b dataset and index have been 
[taken offline](https://laion.ai/notes/laion-maintanence/) (temporarily) and there 
[aren't any alternatives](https://github.com/rom1504/clip-retrieval/issues/324). This is
why we built an index for the Datacomp-12M dataset. While it is a lot smaller than LAION-5b, it
should already enable a lot of use cases again, and can hopefully be the start towards building
indices for more and larger datasets.

- **License:** cc-by-4.0

### Dataset Sources

<!-- Provide the basic links for the dataset. -->

- **Original data:** [datacomp_small](https://huggingface.co/datasets/mlfoundations/datacomp_small)
- **Repository:** [fondant-clip-index](https://github.com/ml6team/fondant-clip-index)

## Uses

<!-- Address questions around how the dataset is intended to be used. -->

We provide an [example use case](https://github.com/ml6team/fondant-usecase-controlnet) which uses the FAISS index of this dataset to create a dataset of interior design images, used for the fine-tuning of a ControlNet model:

## Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

The data repository is structured as follows:
- [data/](https://huggingface.co/datasets/fondant-ai/datacomp-small-clip/viewer/embeddings): The dataset 
  containing ids, urls, and CLIP embeddings
- [faiss](https://huggingface.co/datasets/fondant-ai/datacomp-small-clip/blob/main/faiss): 
  The faiss index
- [id_mapping/](https://huggingface.co/datasets/fondant-ai/datacomp-small-clip/viewer/id_mapping): 
  The mapping of the faiss ids to the original urls

## Dataset Creation

We leveraged Fondant to generate the CLIP index and published the pipeline as a 
[git repository](https://github.com/ml6team/fondant-clip-index). The pipeline consists of 4 steps:

- A [`load_from_hf_hub`](https://fondant.ai/en/stable/components/hub/#load_from_hf_hub#description)
  operation that loads the
  [datacomp_small](https://huggingface.co/datasets/mlfoundations/datacomp_small) dataset from
  huggingface into the Fondant workspace and format.
- A [`download_images`](https://fondant.ai/en/stable/components/hub/#download_images#description)
  operation which downloads the actual images from the urls in the dataset.
- A [`embed_images`](https://fondant.ai/en/stable/components/hub/#embed_images#description) operation which embeds the downloaded images using a CLIP model.
- A [`write_to_file`](https://fondant.ai/en/stable/components/hub/#write_to_file#description)
  operation which writes the original urls and generated embeddings to the chosen destination.

After running the pipeline, we used [`autofaiss`](https://github.com/criteo/autofaiss) to build the
CLIP index.

### Execution details

### Download images

We downloaded the images with 32 cores in parallel, each opening up to 25 concurrent connections,
and achieved a success rate of 72%, resulting in 9.251.172 images.

The downloading was executed on a VM on GCP using the Fondant Docker runner. We originally
planned to run this on Vertex AI, but moved to a VM when noticing lower network bandwidth on Vertex.

The success rate can probably be further improved by setting up a faster DNS resolver.

### Embed images

We leveraged the
[`laion/CLIP-ViT-B-32-laion2B-s34B-b79K`](https://huggingface.co/laion/CLIP-ViT-B-32-laion2B-s34B-b79K)
CLIP model. We chose this model because of a couple of reasons. It is popular, which makes it
easy to use with existing embeddings, it is small, which makes it cheap to run, and it is an open
model trained on open data.

We appreciate any feedback on our choice of model, so we can take this into account if we
generate indices for larger datasets in the future.

The embedding was executed on 4 T4 GPUs on Google Cloud using our Vertex AI runner, with a batch
size of 32. The execution took 8:15 hours.

## Terms and Conditions
Under no circumstances can Fondant be held liable by a third party for (i) the accuracy or correctness of the content, (ii) an alleged infringement of intellectual property rights or (iii) any other alleged claim, action, injunction or suit resulting from the publication or use of the dataset.

## Dataset Card Contact
- Email: [[email protected]](mailto:[email protected])
- Discord: [https://discord.gg/HnTdWhydGp](https://discord.gg/HnTdWhydGp)