Benchmarking Robustness of Adaptation Methods on Pre-trained Vision-Language Models
This work addresses the problem of evaluating robustness for adaptation methods in vision-language models, which is incremental as it benchmarks existing methods rather than proposing new ones.
The study assessed the robustness of 11 adaptation methods for pre-trained vision-language models against multimodal corruptions, finding that they are more sensitive to text corruptions, adapters can outperform full fine-tuning in robustness, and increasing adaptation data or parameters may reduce robustness.
Various adaptation methods, such as LoRA, prompts, and adapters, have been proposed to enhance the performance of pre-trained vision-language models in specific domains. The robustness of these adaptation methods against distribution shifts have not been studied. In this study, we assess the robustness of 11 widely-used adaptation methods across 4 vision-language datasets under multimodal corruptions. Concretely, we introduce 7 benchmark datasets, including 96 visual and 87 textual corruptions, to investigate the robustness of different adaptation methods, the impact of available adaptation examples, and the influence of trainable parameter size during adaptation. Our analysis reveals that: 1) Adaptation methods are more sensitive to text corruptions than visual corruptions. 2) Full fine-tuning does not consistently provide the highest robustness; instead, adapters can achieve better robustness with comparable clean performance. 3) Contrary to expectations, our findings indicate that increasing the number of adaptation data and parameters does not guarantee enhanced robustness; instead it results in even lower robustness. We hope this study could benefit future research in the development of robust multimodal adaptation methods. The benchmark, code, and dataset used in this study can be accessed at https://adarobustness.github.io .