Summary of Multi-scale Multi-instance Visual Sound Localization and Segmentation, by Shentong Mo et al.
Multi-scale Multi-instance Visual Sound Localization and Segmentation
by Shentong Mo, Haofan Wang
First submitted to arxiv on: 31 Aug 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG); Multimedia (cs.MM); Sound (cs.SD); Audio and Speech Processing (eess.AS)
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 A novel multi-scale multi-instance visual sound localization framework, dubbed M2VSL, is introduced to predict the location of objects corresponding to sound sources in videos. Unlike previous methods, which rely on audio-visual associations between global audio and one-scale visual features, M2VSL directly learns multi-scale semantic features associated with sound sources from input images. This is achieved through a learnable multi-scale visual feature extractor and a novel multi-scale multi-instance transformer that dynamically aggregates cross-modal representations for localization. The proposed framework outperforms existing methods on VGGSound-Instruments, VGG-Sound Sources, and AVSBench benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to figure out where things are making sounds in videos is being developed. Right now, computers have trouble matching sounds with the objects that make them in each video frame. To solve this problem, researchers created a new system called M2VSL that can learn many different scales of visual features and match them with sound sources. This means it’s better at finding where things are making sounds in videos than previous methods. |
Keywords
» Artificial intelligence » Transformer
