Hyperbolic Image-text Representations

Karan Desai, Maximilian Nickel, Tanmay Rajpurohit, Justin Johnson, Shanmukha Ramakrishna Vedantam
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:7694-7731, 2023.

Abstract

Visual and linguistic concepts naturally organize themselves in a hierarchy, where a textual concept "dog" entails all images that contain dogs. Despite being intuitive, current large-scale vision and language models such as CLIP do not explicitly capture such hierarchy. We propose MERU, a contrastive model that yields hyperbolic representations of images and text. Hyperbolic spaces have suitable geometric properties to embed tree-like data, so MERU can better capture the underlying hierarchy in image-text datasets. Our results show that MERU learns a highly interpretable and structured representation space while being competitive with CLIP’s performance on standard multi-modal tasks like image classification and image-text retrieval.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-desai23a, title = {Hyperbolic Image-text Representations}, author = {Desai, Karan and Nickel, Maximilian and Rajpurohit, Tanmay and Johnson, Justin and Vedantam, Shanmukha Ramakrishna}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {7694--7731}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/desai23a/desai23a.pdf}, url = {https://proceedings.mlr.press/v202/desai23a.html}, abstract = {Visual and linguistic concepts naturally organize themselves in a hierarchy, where a textual concept "dog" entails all images that contain dogs. Despite being intuitive, current large-scale vision and language models such as CLIP do not explicitly capture such hierarchy. We propose MERU, a contrastive model that yields hyperbolic representations of images and text. Hyperbolic spaces have suitable geometric properties to embed tree-like data, so MERU can better capture the underlying hierarchy in image-text datasets. Our results show that MERU learns a highly interpretable and structured representation space while being competitive with CLIP’s performance on standard multi-modal tasks like image classification and image-text retrieval.} }
Endnote
%0 Conference Paper %T Hyperbolic Image-text Representations %A Karan Desai %A Maximilian Nickel %A Tanmay Rajpurohit %A Justin Johnson %A Shanmukha Ramakrishna Vedantam %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-desai23a %I PMLR %P 7694--7731 %U https://proceedings.mlr.press/v202/desai23a.html %V 202 %X Visual and linguistic concepts naturally organize themselves in a hierarchy, where a textual concept "dog" entails all images that contain dogs. Despite being intuitive, current large-scale vision and language models such as CLIP do not explicitly capture such hierarchy. We propose MERU, a contrastive model that yields hyperbolic representations of images and text. Hyperbolic spaces have suitable geometric properties to embed tree-like data, so MERU can better capture the underlying hierarchy in image-text datasets. Our results show that MERU learns a highly interpretable and structured representation space while being competitive with CLIP’s performance on standard multi-modal tasks like image classification and image-text retrieval.
APA
Desai, K., Nickel, M., Rajpurohit, T., Johnson, J. & Vedantam, S.R.. (2023). Hyperbolic Image-text Representations. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:7694-7731 Available from https://proceedings.mlr.press/v202/desai23a.html.

Related Material