TY - GEN
T1 - On the Test-Time Zero-Shot Generalization of Vision-Language Models
T2 - 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024
AU - Zanella, Maxime
AU - Ayed, Ismail Ben
N1 - Publisher Copyright:
© 2024 IEEE.
PY - 2024
Y1 - 2024
N2 - The development of large vision-language models, notably CLIP, has catalyzed research into effective adaptation techniques, with a particular focus on soft prompt tuning. Conjointly, test-time augmentation, which utilizes multiple augmented views of a single image to enhance zero-shot generalization, is emerging as a significant area of interest. This has predominantly directed research efforts toward test-time prompt tuning. In contrast, we introduce a robust MeanShift for Test-time Augmentation (MTA), which surpasses prompt-based methods without requiring this intensive training procedure. This positions MTA as an ideal solution for both standalone and API-based applications. Additionally, our method does not rely on ad hoc rules (e.g., confidence threshold) used in some previous test-time augmentation techniques to filter the augmented views. Instead, MTA incorporates a quality assessment variable for each view directly into its optimization process, termed as the inlierness score. This score is jointly optimized with a density mode seeking process, leading to an efficient trainingand hyperparameter-free approach. We extensively benchmark our method on 15 datasets and demonstrate MTA's superiority and computational efficiency. Deployed easily as plug-and-play module on top of zero-shot models and state-of-the-art few-shot methods, MTA shows systematic and consistent improvements.
AB - The development of large vision-language models, notably CLIP, has catalyzed research into effective adaptation techniques, with a particular focus on soft prompt tuning. Conjointly, test-time augmentation, which utilizes multiple augmented views of a single image to enhance zero-shot generalization, is emerging as a significant area of interest. This has predominantly directed research efforts toward test-time prompt tuning. In contrast, we introduce a robust MeanShift for Test-time Augmentation (MTA), which surpasses prompt-based methods without requiring this intensive training procedure. This positions MTA as an ideal solution for both standalone and API-based applications. Additionally, our method does not rely on ad hoc rules (e.g., confidence threshold) used in some previous test-time augmentation techniques to filter the augmented views. Instead, MTA incorporates a quality assessment variable for each view directly into its optimization process, termed as the inlierness score. This score is jointly optimized with a density mode seeking process, leading to an efficient trainingand hyperparameter-free approach. We extensively benchmark our method on 15 datasets and demonstrate MTA's superiority and computational efficiency. Deployed easily as plug-and-play module on top of zero-shot models and state-of-the-art few-shot methods, MTA shows systematic and consistent improvements.
KW - CLIP
KW - test-time augmentation
KW - training-free
KW - vision-language
KW - zero-shot
UR - https://www.scopus.com/pages/publications/85202359470
U2 - 10.1109/CVPR52733.2024.02245
DO - 10.1109/CVPR52733.2024.02245
M3 - Contribution to conference proceedings
AN - SCOPUS:85202359470
T3 - Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
SP - 23783
EP - 23793
BT - Proceedings - 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024
PB - IEEE Computer Society
Y2 - 16 June 2024 through 22 June 2024
ER -