Query-centric Audio-Visual Cognition Network for Moment Retrieval, Segmentation and Step-Captioning

Authors: Yunbin Tu, Liang Li, Li Su, Qingming Huang

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show QUAG achieves the SOTA results on HIREST. Further, we test QUAG on the query-based video summarization task and verify its good generalization.
Researcher Affiliation Academia Yunbin Tu1, Liang Li2,1*, Li Su1,3*, Qingming Huang1 1 School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing, China 2 Key Laboratory of AI Safety of CAS, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China 3Peng Cheng Laboratory, Shenzhen, China EMAIL, EMAIL, EMAIL
Pseudocode No The paper describes the methodology using narrative text and mathematical formulations (e.g., equations 1-15) but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code https://github.com/tuyunbin/QUAG
Open Datasets Yes HIREST consists of the tasks of video retrieval, moment retrieval, moment segmentation, and step-captioning. It is comprised of 3.4K text-video pairs, 1.8K moments, and 8.6K step captions. We use the official split with 1,507 video-query pairs for training, 477 video-query pairs for validation and 1,391 video-query pairs for testing.
Dataset Splits Yes HIREST: We use the official split with 1,507 video-query pairs for training, 477 video-query pairs for validation and 1,391 video-query pairs for testing. TVSum: For a fair-comparison, we follow QDDETR (Moon et al. 2023) to utilize 80% videos for training and the remaining for testing.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory used for running the experiments. It only mentions using pre-trained models and fine-tuning.
Software Dependencies No The paper lists several software tools and models used (e.g., EVA-CLIP, Whisper, Mini LM, CLIP4Caption, AdamW optimizer) but does not provide specific version numbers for these software dependencies, which is required for reproducible description.
Experiment Setup Yes The hidden size is set to 768. During training, the batch size is set to 5 and learning rate is set to 1e-5, Adam W optimizer (Loshchilov and Hutter 2018) is used to minimize the training loss defined in Eq. (15).