Summary of Label-anticipated Event Disentanglement For Audio-visual Video Parsing, by Jinxing Zhou et al.
Label-anticipated Event Disentanglement for Audio-Visual Video Parsing
by Jinxing Zhou, Dan Guo, Yuxin Mao, Yiran Zhong, Xiaojun Chang, Meng Wang
First submitted to arxiv on: 11 Jul 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The Audio-Visual Video Parsing (AVVP) task aims to identify and temporally locate events within both audio and visual modalities. Traditional methods focus on improving early encoders, but the decoding phase often receives less attention. This paper introduces a new decoding paradigm called LEAP, which employs label texts of event categories to parse potentially overlapping events. The process iteratively projects encoded features onto semantically independent label embeddings, disentangling event semantics and refining relevant label embeddings for a more discriminative and interpretable decoding process. LEAP is facilitated by a semantic-aware optimization strategy that includes an audio-visual semantic similarity loss function. This novel metric uses the Intersection over Union of audio and visual events (EIoU) to calibrate audio-visual similarities at the feature level, accommodating varied event densities across modalities. The paper demonstrates the superiority of LEAP with state-of-the-art performance for AVVP and enhanced audio-visual event localization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary AVVP is a task that aims to identify and locate events within both audio and visual modalities. This can be challenging because multiple events can overlap in time. Traditional methods focus on improving early encoders, but the decoding phase often receives less attention. The paper introduces a new way of decoding called LEAP, which uses label texts to help parse overlapping events. LEAP works by projecting features onto semantically independent label embeddings and refining relevant label embeddings. The paper also proposes an optimization strategy that includes a novel metric for calibrating audio-visual similarities. This helps accommodate varied event densities across modalities. The results show that LEAP outperforms other methods, achieving state-of-the-art performance for AVVP and enhancing the relevant task of locating events in audio and visual data. |
Keywords
» Artificial intelligence » Attention » Loss function » Optimization » Parsing » Semantics
