Text-Guided Image Clustering

Authors: Andreas Stephan, Lukas Miklautz, Kevin Sidak, Jan Philip Wahle, Bela Gipp, Claudia Plant, Benjamin Roth

This repo contains the code related to the paper "Text-Guided Image Clustering", accepted to the main conference at EACL 2024.

If there's questions, please contact us at andreas.stephan@univie.ac.at

---

Abstract

Image clustering divides a collection of images into meaningful groups, typically interpreted post-hoc via human-given annotations. Those are usually in the form of text, begging the question of using text as an abstraction for image clustering. Current image clustering methods, however, neglect the use of generated textual descriptions. We, therefore, propose \textit{Text-Guided Image Clustering}, i.e. generating text using image captioning and visual question-answering (VQA) models, and subsequently clustering the generated text. Further, we introduce a novel approach to inject task- or domain knowledge for clustering by prompting VQA models. Across eight diverse image clustering datasets, our results show that the obtained text representations often outperform image features. Additionally, we propose a counting-based cluster explainability method. Our evaluations show that the derived keyword-based explanations describe clusters better than the respective cluster accuracy suggests. Overall, this research challenges traditional approaches and paves the way for a paradigm shift in image clustering, using generated text.

---

Installation

There are two requirement files. If you want to explicitly work with our versions, run pip install fixed_requirements.txt In case you want to work with newer version of the used libraries, run pip install requirements.txt. Note that we do not continuously test whether this works.

Finally you can run pip install -e . to make sure imports work.

Usage

If anything is unclear, don't hesitate, contact us immediately andreas.stephan@univie.ac.at.

Image-to-text

In order to run captioning, there is an example config in caption/blip2_vqa/config.cfg, so you can run it via

PYTHONPATH=. python caption/blip2.py  --flagfile=examples/blip2_vqa/config.cfg

Make sure to change the data folder paths first.

SentenceBert

To add SentenceBert embeddings to a dataset, you just need to set the "data_dir" + "load_dir" arguments to the previous "save_dir"

PYTHONPATH=. python embed_sentence_bert.py  --flagfile=examples/blip2_vqa/config.cfg

KMeans Evaluation

PYTHONPATH=. python embed_sentence_bert.py  --flagfile=examples/blip2_vqa/config.cfg

Citation

@inproceedings{stephan-etal-2024-text,
    title = "Text-Guided Image Clustering",
    author = "Stephan, Andreas  and
      Miklautz, Lukas  and
      Sidak, Kevin  and
      Wahle, Jan Philip  and
      Gipp, Bela  and
      Plant, Claudia  and
      Roth, Benjamin",
    editor = "Graham, Yvette  and
      Purver, Matthew",
    booktitle = "Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = mar,
    year = "2024",
    address = "St. Julian{'}s, Malta",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.eacl-long.180",
    pages = "2960--2976",
    abstract = "Image clustering divides a collection of images into meaningful groups, typically interpreted post-hoc via human-given annotations. Those are usually in the form of text, begging the question of using text as an abstraction for image clustering. Current image clustering methods, however, neglect the use of generated textual descriptions. We, therefore, propose \textit{Text-Guided Image Clustering}, i.e., generating text using image captioning and visual question-answering (VQA) models and subsequently clustering the generated text. Further, we introduce a novel approach to inject task- or domain knowledge for clustering by prompting VQA models. Across eight diverse image clustering datasets, our results show that the obtained text representations often outperform image features. Additionally, we propose a counting-based cluster explainability method. Our evaluations show that the derived keyword-based explanations describe clusters better than the respective cluster accuracy suggests. Overall, this research challenges traditional approaches and paves the way for a paradigm shift in image clustering, using generated text.",
}