Summary of Efficient Multiscale Multimodal Bottleneck Transformer For Audio-video Classification, by Wentao Zhu
Efficient Multiscale Multimodal Bottleneck Transformer for Audio-Video Classification
by Wentao Zhu
First submitted to arxiv on: 8 Jan 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); 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 This research proposes a novel neural network architecture called Multiscale Multimodal Transformer (MMT) that effectively combines audio and video signals to improve action recognition. The MMT model consists of two components: Multiscale Audio Transformer (MAT) and Multiscale Video Transformer, which learn hierarchical representations from both modalities. To align the audio and video streams, the authors introduce multimodal contrastive objectives, including audio-video contrastive loss (AVC) and intra-modal contrastive loss (IMC). The proposed model surpasses previous state-of-the-art approaches on Kinetics-Sounds and VGGSound datasets without using external training data. Additionally, the MAT component outperforms a baseline AudioStreamTransformer (AST) on three public benchmark datasets while being more efficient in terms of computational resources. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research creates a new way to use both sound and video together to recognize actions. The idea is called Multiscale Multimodal Transformer (MMT). MMT has two parts: one for audio and one for video. It learns to represent these signals in different ways, which helps it understand the relationship between what we hear and see. To make this connection stronger, the researchers developed special techniques that align the audio and video streams. This new approach works better than previous methods without needing extra training data. The part of MMT focused on audio is also more effective and efficient than a similar existing model. |
Keywords
* Artificial intelligence * Contrastive loss * Neural network * Transformer
