One Single Hub Text Breaks CLIP: Identifying Vulnerabilities in Cross-Modal Encoders via Hubness
About
The hubness problem, in which hub embeddings are close to many unrelated examples, occurs often in high-dimensional embedding spaces and may pose a practical threat for purposes such as information retrieval and automatic evaluation metrics. In particular, since cross-modal similarity between text and images cannot be calculated by direct comparisons, such as string matching, cross-modal encoders that project different modalities into a shared space are helpful for various cross-modal applications, and thus, the existence of hubs may pose practical threats. To reveal the vulnerabilities of cross-modal encoders, we propose a method for identifying the hub embedding and its corresponding hub text. Experiments on image captioning evaluation in MSCOCO and nocaps along with image-to-text retrieval tasks in MSCOCO and Flickr30k showed that our method can identify a single hub text that unreasonably achieves comparable or higher similarity scores than human-written reference captions in many images, thereby revealing the vulnerabilities in cross-modal encoders.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Text-to-Image Alignment | MS-COCO | CLIP Score1.023 | 60 | |
| Image-Text Alignment Evaluation | NoCaps Out-of-Domain (val) | CLIPScore84.1 | 40 | |
| Caption Evaluation | MSCOCO | Win Rate90 | 20 | |
| Caption Evaluation | NoCaps | Win Rate71.1 | 20 | |
| Image Captioning | MSCOCO | CLIPScore40.6 | 14 |