SSGAN: A Semantic Similarity-Based GAN for Small-Sample Image Augmentation

Congcong Ma,Jiaqi Mi,Wanlin Gao,Sha Tao

DOI: https://doi.org/10.1007/s11063-024-11498-z

IF: 2.565

2024-04-17

Neural Processing Letters

Abstract:Image sample augmentation refers to strategies for increasing sample size by modifying current data or synthesizing new data based on existing data. This technique is of vital significance in enhancing the performance of downstream learning tasks in widespread small-sample scenarios. In recent years, GAN-based image augmentation methods have gained significant attention and research focus. They have achieved remarkable generation results on large-scale datasets. However, their performance tends to be unsatisfactory when applied to datasets with limited samples. Therefore, this paper proposes a semantic similarity-based small-sample image augmentation method named SSGAN. Firstly, a relatively shallow pyramid-structured GAN-based backbone network was designed, aiming to enhance the model's feature extraction capabilities to adapt to small sample sizes. Secondly, a feature selection module based on high-dimensional semantics was designed to optimize the loss function, thereby improving the model's learning capacity. Lastly, extensive comparative experiments and comprehensive ablation experiments were carried out on the "Flower" and "Animal" datasets. The results indicate that the proposed method outperforms other classical GANs methods in well-established evaluation metrics such as FID and IS, with improvements of 18.6 and 1.4, respectively. The dataset augmented by SSGAN significantly enhances the performance of the classifier, achieving a 2.2% accuracy improvement compared to the best-known method. Furthermore, SSGAN demonstrates excellent generalization and robustness.

computer science, artificial intelligence

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the issues of overfitting and limited generalization capability in deep learning tasks caused by small sample image datasets. Specifically, the paper proposes a small sample image augmentation method based on semantic similarity (SSGAN) to improve the performance of existing Generative Adversarial Networks (GANs) on small sample datasets. #### Main Contributions and Innovations: 1. **Pyramid Structure Backbone Network**: Designed a relatively shallow pyramid structure GAN backbone network to enhance the model's ability to extract features from small sample images. 2. **High-Dimensional Semantic Feature Selection Module**: Introduced a feature selection module based on high-dimensional semantic features to optimize the loss function, thereby improving the model's learning capability. 3. **Extensive Experimental Validation**: Conducted extensive comparative experiments and detailed ablation studies on the "Flower" and "Animal" datasets, demonstrating the effectiveness of SSGAN in small sample augmentation. ### Experimental Results - On the commonly used evaluation metrics FID and IS, SSGAN improved by 18.6 and 1.4 respectively, indicating that the generated images have good clarity and diversity. - The dataset enhanced by SSGAN significantly improved the performance of the classifier, with a 2.2% increase in accuracy compared to the best-known methods. Through these improvements and experimental validations, the paper demonstrates the superior performance of SSGAN in small sample image augmentation.

SSGAN: A Semantic Similarity-Based GAN for Small-Sample Image Augmentation

WGAN-CL: A Wasserstein GAN with confidence loss for small-sample augmentation

Stacked Siamese Generative Adversarial Nets: A Novel Way to Enlarge Image Dataset

Pixel Level Data Augmentation for Semantic Image Segmentation Using Generative Adversarial Networks.

Semantic feature augmentation for fine-grained visual categorization with few-sample training

Local and Global GANs with Semantic-Aware Upsampling for Image Generation

Vsa-Cgan: An Intelligent Generation Model For Deep Learning Sample Database Construction

Image Augmentations for GAN Training

Differentiable Augmentation for Data-Efficient GAN Training

Data Augmentation Based on Generative Adversarial Network with Mixed Attention Mechanism

SGBGAN: minority class image generation for class-imbalanced datasets

Semantic Image Analogy with a Conditional Single-Image GAN

Improving GAN Training via Feature Space Shrinkage

ESGAN for Generating High Quality Enhanced Samples

Semantic-Guided Generative Image Augmentation Method with Diffusion Models for Image Classification

Small facial image dataset augmentation using conditional GANs based on incomplete edge feature input

Optimizing Data Augmentation for Semantic Segmentation on Small-Scale Dataset

VSGAN: Visual Saliency Guided Generative Adversarial Network for Data Augmentation

A survey on Image Data Augmentation for Deep Learning

Conditional Generative Adversarial Network-Based Training Sample Set Improvement Model for the Semantic Segmentation of High-Resolution Remote Sensing Images

DESRGAN: Detail-Enhanced Generative Adversarial Networks for Small Sample Single Image Super-Resolution