Understanding the Vulnerability of CLIP to Image Compression
This work addresses a robustness issue for users of CLIP and similar vision-language models, but it is incremental as it focuses on analyzing an existing vulnerability rather than proposing a new solution.
The paper tackles the problem of CLIP's vulnerability to image compression, showing that compression significantly reduces its zero-shot recognition accuracy, with evaluations on CIFAR-10 and STL-10 providing concrete evidence of this effect.
CLIP is a widely used foundational vision-language model that is used for zero-shot image recognition and other image-text alignment tasks. We demonstrate that CLIP is vulnerable to change in image quality under compression. This surprising result is further analysed using an attribution method-Integrated Gradients. Using this attribution method, we are able to better understand both quantitatively and qualitatively exactly the nature in which the compression affects the zero-shot recognition accuracy of this model. We evaluate this extensively on CIFAR-10 and STL-10. Our work provides the basis to understand this vulnerability of CLIP and can help us develop more effective methods to improve the robustness of CLIP and other vision-language models.