In recent years, Cross-Modal Retrieval (CMR), which can retrieve data across types based on query semantics, has become an attractive technology due to the widespread applications of multimedia data. Outsourcing multimedia data to a cloud server is a reliable way to improve the quality of CMR services, but it will also incur potential data privacy leakage issues. Existing schemes for privacy-preserving outsourced data search services are either inapplicable to CMR or limited by efficiency and scalability. To address the above issues, we investigate the problem of Searchable Symmetric Encryption (SSE) for CMR in this paper. Firstly, we formulate the definition of SSE for CMR (namely, SSECMR) and extend the SSE leakage functions to capture the leakage in SSECMR. Then, by constructing distance-computation-free Hamming inverted multi-index, we propose a practical SSECMRconstruction. Specifically, we transform Hamming distance-based range queries into multi-key queries, thereby avoiding computationally expensive comparison operations and enabling efficient Hamming distance queries over encrypted data. Our design supports secure CMR with sub-linear complexity in a single communication roundtrip. Through rigorous security analysis, we demonstrate that our construction can provide adaptive security. Empirical evaluations on real-world datasets demonstrate that SSECMRoutperforms the state-of-the-art scheme in both efficiency and accuracy, and is comparable to plaintext applications. Our code is available at https://github.com/SSE-CMR/SSECMR.

Benchmarks for large multimodal language models (MLMs) now serve to simultaneously assess the general capabilities of models instead of evaluating for a specific capability. As a result, when a developer wants to identify which models to use for their application, they are overwhelmed by the number of benchmarks and remain uncertain about which benchmark's results are most reflective of their specific use case. This paper introduces Task-Me-Anything, a benchmark generation engine which produces a benchmark tailored to a user's needs. Task-Me-Anything maintains an extendable taxonomy of visual assets and can programmatically generate a vast number of task instances. Additionally, it algorithmically addresses user queries regarding MLM performance efficiently within a computational budget. It contains 113K images, 10K videos, 2K 3D object assets, over 365 object categories, 655 attributes, and 335 relationships. It can generate 500M image/video question-answering pairs, which focus on evaluating MLM perceptual capabilities. Task-Me-Anything reveals critical insights: open-source MLMs excel in object and attribute recognition but lack spatial and temporal understanding; each model exhibits unique strengths and weaknesses; larger models generally perform better, though exceptions exist; and GPT4O demonstrates challenges in recognizing rotating/moving objects and distinguishing colors.