Learning with privileged information via adversarial discriminative modality distillation

Abstract

Heterogeneous data modalities can provide complementary cues for several tasks, usually leading to more robust algorithms and better performance. However, while training data can be accurately collected to include a variety of sensory modalities, it is often the case that not all of them are available in real life (testing) scenarios, where a model has to be deployed. This raises the challenge of how to extract information from multimodal data in the training stage, in a form that can be exploited at test time, considering limitations such as noisy or missing modalities. This paper presents a new approach in this direction for RGB-D vision tasks, developed within the adversarial learning and privileged information frameworks. We consider the practical case of learning representations from depth and RGB videos, while relying only on RGB data at test time. We propose a new approach to train a hallucination network that learns to distill depth information via adversarial learning, resulting in a clean approach without several losses to balance or hyperparameters. We report state-of-the-art results on object classification on the NYUD dataset and video action recognition on the largest multimodal dataset available for this task, the NTU RGB+D, as well as on the Northwestern-UCLA.

arXiv commentAccepted to IEEE Transactions on Pattern Analysis and Machine Intelligence

References in corpus (7)

• Distilling the Knowledge in a Neural Network
• Conditional Generative Adversarial Nets
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Two-Stream Convolutional Networks for Action Recognition in Videos
• Unifying distillation and privileged information
• Deep CNN Ensemble with Data Augmentation for Object Detection
• Viewpoint Invariant Action Recognition using RGB-D Videos

Cited by in corpus (9)

• Human Action Recognition from Various Data Modalities: A Review
• Semantics-aware Adaptive Knowledge Distillation for Sensor-to-Vision Action Recognition
• DMCL: Distillation Multiple Choice Learning for Multimodal Action Recognition
• Flow-Distilled IP Two-Stream Networks for Compressed Video Action Recognition
• VideoSSL: Semi-Supervised Learning for Video Classification
• 3D-to-2D Distillation for Indoor Scene Parsing
• KAN-HyperpointNet for Point Cloud Sequence-Based 3D Human Action Recognition
• Cross-Modal Knowledge Transfer via Inter-Modal Translation and Alignment for Affect Recognition
• Feature-Supervised Action Modality Transfer