Supplementary Materials for MUVR: AMulti-Modal Untrimmed Video Retrieval Benchmark with Multi-Level Visual Correspondence Anonymous Author(s) Affiliation Address email

In this supplementary material, we elaborate on the MLLMs prompting details in Section 1. We1 further illustrate the annotation instructions in Section 2. Then, some visualization examples are2 provided in Section 3. Limitations and social impact are introduced in Section 4.3 The evaluation prompts for MLLMs are listed in Table 1 and 2. Although we attempted to maintain5 consistency across models, slight variations were necessary due to differing prompting requirements.6 We take the relationship annotation of9 the News partition as an example, while other partitions have different visual correspondences.10 3 Visualization11 Figure 1, 2, 3, 4 and 5 provide several relevant examples of different partitions from MUVR, with a12 text description of the query video and the tag of each video.13 MUVR relies on human annotators to annotate videos with rich semantics.