GAN Framework Research Articles

GAN-Poser: an improvised bidirectional GAN model for human motion prediction

A novel method called GAN-Poser has been explored to predict human motion in less time given an input 3D human skeleton sequence based on a generator–discriminator framework. Specifically, rather than using the conventional Euclidean loss, a frame-wise geodesic loss is used for geometrically meaningful and more precise distance measurement. In this paper, we have used a bidirectional GAN framework along with a recursive prediction strategy to avoid mode-collapse and to further regularize the training. To be able to generate multiple probable human-pose sequences conditioned on a given starting sequence, a random extrinsic factor $$\varTheta$$ has also been introduced. The discriminator is trained in order to regress the extrinsic factor $$\varTheta$$ , which is used alongside with the intrinsic factor (encoded starting pose sequence) to generate a particular pose sequence. In spite of being in a probabilistic framework, the modified discriminator architecture allows predictions of an intermediate part of pose sequence to be used as conditioning for prediction of the latter part of the sequence. This adversarial learning-based model takes into consideration of the stochasticity, and the bidirectional setup provides a new direction to evaluate the prediction quality against a given test sequence. Our resulting novel method, GAN-Poser, achieves superior performance over the state-of-the-art deep learning approaches when evaluated on the standard NTU-RGB-D and Human3.6 M dataset.

DeGAN: Data-Enriching GAN for Retrieving Representative Samples from a Trained Classifier

In this era of digital information explosion, an abundance of data from numerous modalities is being generated as well as archived everyday. However, most problems associated with training Deep Neural Networks still revolve around lack of data that is rich enough for a given task. Data is required not only for training an initial model, but also for future learning tasks such as Model Compression and Incremental Learning. A diverse dataset may be used for training an initial model, but it may not be feasible to store it throughout the product life cycle due to data privacy issues or memory constraints. We propose to bridge the gap between the abundance of available data and lack of relevant data, for the future learning tasks of a given trained network. We use the available data, that may be an imbalanced subset of the original training dataset, or a related domain dataset, to retrieve representative samples from a trained classifier, using a novel Data-enriching GAN (DeGAN) framework. We demonstrate that data from a related domain can be leveraged to achieve state-of-the-art performance for the tasks of Data-free Knowledge Distillation and Incremental Learning on benchmark datasets. We further demonstrate that our proposed framework can enrich any data, even from unrelated domains, to make it more useful for the future learning tasks of a given network.

Unsupervised image translation using adversarial networks for improved plant disease recognition

Acquisition of data in task-specific applications of machine learning like plant disease recognition is a costly endeavor owing to the requirements of professional human diligence and time constraints. In this paper, we present a pipeline that uses GANs in an unsupervised image translation environment to improve learning with respect to the data distribution in a plant disease dataset, reducing the partiality introduced by acute class imbalance and hence shifting the classification decision boundary towards better performance. The empirical analysis of our method is demonstrated on a limited dataset of 2789 tomato plant disease images, highly corrupted with an imbalance in the 9 disease categories. First, we extend the state of the art for the GAN-based image-to-image translation method by enhancing the perceptual quality of the generated images and preserving the semantics. We introduce AR-GAN, where in addition to the adversarial loss, our synthetic image generator optimizes on Activation Reconstruction loss (ARL) function that optimizes feature activations against the natural image. We present visually more compelling synthetic images in comparison to most prominent existing models and evaluate the performance of our GAN framework in terms of various datasets and metrics. Second, we evaluate the performance of a baseline convolutional neural network classifier for improved recognition using the resulting synthetic samples to augment our training set and compare it with the classical data augmentation scheme. We observe a significant improvement in classification accuracy (+5.2%) using generated synthetic samples as compared to (+0.8%) increase using classic augmentation in an equal class distribution environment.

Sparsely Self-Supervised Generative Adversarial Nets for Radio Frequency Estimation

Radio frequency (RF) estimation plays a significant role in cellular network’s planning and optimization. The conventional methods for RF estimation are mainly based on radio propagation models, which suffer from low accuracy and coarse granularity. Distinguished from existing methods, we propose sparsely self-supervised generative adversarial nets (SS-GAN), a novel data-driven model to generate the RF maps of an area from irregularly distributed measurement samples. SS-GAN meticulously adopts the standard GAN framework, where the generator learns the distribution of true observations under the guidance of the discriminator that discriminates whether the input data is from real samples or from generated outputs. Competition in the minmax game between generator and discriminator drives them to improve their capability, until the generator is indistinguishable from the true RF distribution. Specifically, on top of the GAN framework, SS-GAN carries out a variety of operations to enhance the estimation: (1) In addition to observations of the measured RF coverage and RF interference, SS-GAN also employs a collection of crucial auxiliary information (e.g., geographic data) as additional input features to the GAN framework so as to precisely characterize the RF environment; (2) To dampen the training instability, a new lightweight reconstruction loss is introduced to the objective function of SS-GAN rather than solely using the adversarial loss, which aims to impose a weak supervision on the generated RF maps according to estimation accuracy; (3) Moreover, SS-GAN designs an innovative sparsely self-supervised ( ${\mathcal {SS}}$ ) learning mechanism that facilitates the validation of the estimated results for a model lacking direct ground truth knowledge. Extensive experiments on a real-world 4G LTE dataset demonstrate that SS-GAN can substantially improve the estimation accuracy over the state-of-the-art baselines. Comparison results are presented through visualized case studies and quantitative statistics.