Abstract:Videos are inherently multimodal. This paper studies the problem of how to fully exploit the abundant multimodal clues for improved video categorization. We introduce a hybrid deep learning framework that integrates useful clues from multiple modalities, including static spatial appearance information, motion patterns within a short time window, audio information as well as long-range temporal dynamics. More specifically, we utilize three Convolutional Neural Networks (CNNs) operating on appearance, motion and audio signals to extract their corresponding features. We then employ a feature fusion network to derive a unified representation with an aim to capture the relationships among features. Furthermore, to exploit the long-range temporal dynamics in videos, we apply two Long Short Term Memory networks with extracted appearance and motion features as inputs. Finally, we also propose to refine the prediction scores by leveraging contextual relationships among video semantics. The hybrid deep learning framework is able to exploit a comprehensive set of multimodal features for video classification. Through an extensive set of experiments, we demonstrate that (1) LSTM networks which model sequences in an explicitly recurrent manner are highly complementary with CNN models; (2) the feature fusion network which produces a fused representation through modeling feature relationships outperforms alternative fusion strategies; (3) the semantic context of video classes can help further refine the predictions for improved performance. Experimental results on two challenging benchmarks, the UCF-101 and the Columbia Consumer Videos (CCV), provide strong quantitative evidence that our framework achieves promising results: $93.1\%$ on the UCF-101 and $84.5\%$ on the CCV, outperforming competing methods with clear margins.

Exploiting Objects with LSTMs for Video Categorization

Exploiting long-term temporal dynamics for video captioning

Modeling Spatial-Temporal Clues in a Hybrid Deep Learning Framework for Video Classification

Exploiting Temporal Coherence for Multi-modal Video Categorization

Modelling a Spatial-Motion Deep Learning Framework to Classify Dynamic Patterns of Videos

STC: Spatio-Temporal Contrastive Learning for Video Instance Segmentation.

Exploiting Feature and Class Relationships in Video Categorization with Regularized Deep Neural Networks.

Semantics Meets Temporal Correspondence: Self-supervised Object-centric Learning in Videos

Relational Long Short-Term Memory for Video Action Recognition

Exploiting Temporal Coherence for Self-Supervised Visual Tracking by Using Vision Transformer

Effective and Robust: A Discriminative Temporal Learning Transformer for Satellite Videos

Temporal Dynamic Graph LSTM for Action-driven Video Object Detection

Temporally Identity-Aware SSD With Attentional LSTM

Modeling Multimodal Clues in a Hybrid Deep Learning Framework for Video Classification

TS-LSTM and Temporal-Inception: Exploiting Spatiotemporal Dynamics for Activity Recognition

Unsupervised Learning of Video Representations using LSTMs

Learning Spatial-Semantic Features for Robust Video Object Segmentation

Temporal-Spatial Mapping for Action Recognition

CC-LSTM: Cross and Conditional Long-Short Time Memory for Video Captioning

Campus Abnormal Behavior Recognition With Temporal Segment Transformers

Learning Temporal Distribution and Spatial Correlation Toward Universal Moving Object Segmentation