Audio-Visual Model Distillation Using Acoustic Images

By Andrés F. Pérez, Valentina Sanguineti, Pietro Morerio, Vittorio Murino. 2019

31 citations recorded by the source corpus.

In this paper, we investigate how to learn rich and robust feature representations for audio classification from visual data and acoustic images, a novel audio data modality. Former models learn audio representations from raw signals or spectral data acquired by a single microphone, with remarkable results in classification and retrieval. However, such representations are not so robust towards variable environmental sound conditions. We tackle this drawback by exploiting a new multimodal labeled action recognition dataset acquired by a hybrid audio-visual sensor that provides RGB video, raw audio signals, and spatialized acoustic data, also known as acoustic images, where the visual and acoustic images are aligned in space and synchronized in time. Using this richer information, we train audio deep learning models in a teacher-student fashion. In particular, we distill knowledge into audio networks from both visual and acoustic image teachers. Our experiments suggest that the learned representations are more powerful and have better generalization capabilities than the features learned from models trained using just single-microphone audio data.

Subfields: Cognitive Neuroscience, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

1. Jun Liu
2. Dacheng Tao
3. Qiuhong Ke
4. S. Maybank
5. Ming-Hsuan Yang
6. Hossein Rahmani
7. Zehua Sun
8. In-So Kweon

In Signal Processing

1. Jun Liu
2. Dacheng Tao
3. Qiuhong Ke
4. S. Maybank
5. Ming-Hsuan Yang
6. Hossein Rahmani
7. Zehua Sun
8. In-So Kweon

Citations in the corpus

• Knowledge Distillation: A Survey (2020, methodology): “In essence, this distillation shares the cross-modal knowledge among the teachers and students (Perez et al. 2020; Albanie et al. 2018; Roheda et al. 2018).”
• Learning to Localize Sound Sources in Visual Scenes: Analysis and Applications (2019, methodology): “Although the sound localization task is not effectively addressedwith our unsupervised learning approach with static images and mono audios, other methods that use spatial microphones [25], [53], [54], [55] or”
• Human Action Recognition From Various Data Modalities: A Review (2020, background): “There are also several works on co-learning between visual and non-visual modalities [299], [364], [365].”