Abstract:Recognition of activities in the video is an important field in computer vision. Many successful works have been done on activity recognition and they achieved acceptable results in recent years. However, their training is completely static, meaning that all classes are taught to the system in one training step. The system is only able to recognize the equivalent classes. The main disadvantage of this type of training is that if new classes need to be taught to the system, the system must be retrained from scratch and all classes retaught to the system. This specification has many challenges, such as storing and retaining data and respending training costs. We propose an approach for training the action recognition system in video data which can teach new classes to the system without the need for previous data. We will provide an incremental learning algorithm for class recognition tasks in video data. Two different approaches are combined to prevent catastrophic forgetting in the proposed algorithm. In the proposed incremental learning algorithm, two approaches are introduced and used to maintain network information in combination. These two approaches are network sharing and network knowledge distillation. We introduce a neural network architecture for action recognition to understand and represent the video data. We propose the distillation of network knowledge at the classification and feature level, which can be divided into spatial and temporal parts at the feature level. We also suggest initializing new classifiers using previous classifiers. The proposed algorithm is evaluated on the USCF101, HMDB51, and Kinetics-400 datasets. We will consider various factors such as the amount of distillation knowledge, the number of new classes and the incremental learnings stages, and their impact on the final recognition system. Finally, we will show that the proposed algorithm can teach new classes to the recognition system without forgetting the previous classes and does not need the previous data or exemplar data.

Class-Incremental Learning for Action Recognition in Videos

Class-Incremental Learning on Video-Based Action Recognition by Distillation of Various Knowledge

When Video Classification Meets Incremental Classes

Class-Incremental Learning with Multiscale Distillation for Weakly Supervised Temporal Action Localization.

Continual Recognition with Adaptive Memory Update.

Hypercorrelation Evolution for Video Class-Incremental Learning

Incremental Learning of Object Detectors without Catastrophic Forgetting

Class-Incremental Learning: Survey and Performance Evaluation on Image Classification

Model Behavior Preserving for Class-Incremental Learning

Class-Incremental Learning: A Survey

Class Incremental Learning for Video Action Classification

Class-Incremental Learning via Knowledge Amalgamation

Rethinking Class-Incremental Learning from a Dynamic Imbalanced Learning Perspective

Class Incremental Learning with Self-Supervised Pre-Training and Prototype Learning

Towards Class-incremental Object Detection with Nearest Mean of Exemplars

Incremental Learning for Online Data Using QR Factorization on Convolutional Neural Networks

Incremental class learning using variational autoencoders with similarity learning

MCF-VC: Mitigate Catastrophic Forgetting in Class-Incremental Learning for Multimodal Video Captioning

An Efficient Replay for Class-Incremental Learning with Pre-trained Models

On the Exploration of Incremental Learning for Fine-grained Image Retrieval