TrustCLIP: Learning Private Visual Features via Adversarial Reconstruction
About
Vision and vision-language models rely on high-level visual representations that are increasingly used across recognition, retrieval, and multimodal reasoning pipelines. However, recent advances in generative modeling have shown that such features can often be inverted, enabling realistic reconstructions of the underlying image and raising significant privacy risks. We revisit this problem through the lens of reconstruction and propose TrustCLIP, a reconstruction-driven framework that treats a feature-conditioned generator as an explicit privacy adversary. TrustCLIP learns a projection between encoder features and downstream modules that is explicitly optimized to degrade the reconstructions produced by generative attackers while retaining the necessary signals for downstream tasks. Unlike prior defenses that rely on discriminative privacy metrics, TrustCLIP directly optimizes against a generative reconstruction attacker, targeting a threat not captured by standard evaluation protocols. We demonstrate its effectiveness in both conventional classification and multimodal large language model pipelines. Across these settings, TrustCLIP consistently reduces the fidelity of generative inversions while maintaining downstream task performance. Project page: https://atnikos.github.io/trustclip/
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Multimodal Evaluation | MM-Vet | -- | 249 | |
| Object Hallucination Evaluation | POPE | Accuracy (POPE)86.1 | 137 | |
| Multi-modal Perception Evaluation | MME Perception | Perception Score1.39e+3 | 43 | |
| Multimodal Evaluation | LLaVA-Bench-Wild (LLaVA-W) | Overall Score65.3 | 43 | |
| Visual Question Answering | ScienceQA (SQAI) | Accuracy62.1 | 42 | |
| Visual Question Answering | TextVQA | VQAText Score55 | 21 | |
| Visual Question Answering | VQA v2 | VQAv2 Score76.3 | 16 | |
| Multimodal Evaluation | SEED-Bench Image | SEEDI Score63 | 7 | |
| Image Reconstruction | Reconstruction Quality under IP-Adapter Attack | PSNR10.42 | 3 |