DiscrimNet: Semi-Supervised Action Recognition from Videos using Generative Adversarial Networks

TL;DR

DiscrimNet introduces a semi-supervised action recognition method that leverages GANs by pre-training a discriminator on unlabeled videos and fine-tuning on labeled data, achieving competitive results.

Contribution

The paper presents a novel semi-supervised framework using GAN discriminators for action recognition from videos, with optimized network settings for improved performance.

Findings

Achieves superior or comparable performance to state-of-the-art methods on UCF101 and HMDB51 datasets.

Utilizes only appearance information for action recognition.

Demonstrates effective use of GAN discriminator as a feature extractor.

Abstract

We propose an action recognition framework using Gen- erative Adversarial Networks. Our model involves train- ing a deep convolutional generative adversarial network (DCGAN) using a large video activity dataset without la- bel information. Then we use the trained discriminator from the GAN model as an unsupervised pre-training step and fine-tune the trained discriminator model on a labeled dataset to recognize human activities. We determine good network architectural and hyperparameter settings for us- ing the discriminator from DCGAN as a trained model to learn useful representations for action recognition. Our semi-supervised framework using only appearance infor- mation achieves superior or comparable performance to the current state-of-the-art semi-supervised action recog- nition methods on two challenging video activity datasets: UCF101 and HMDB51.