EqGAN: Feature Equalization Fusion for Few-shot Image Generation

TL;DR

EqGAN introduces a novel feature equalization fusion approach in GANs to enhance quality and diversity in few-shot image generation by disentangling and aligning structure and texture features across scales.

Contribution

The paper proposes a new fusion strategy with separate branches for structure and texture, and a consistent equalization loss to improve few-shot image generation quality.

Findings

Significantly improves FID and LPIPS scores

Outperforms state-of-the-art methods in generation quality

Enhances downstream classification accuracy

Abstract

Due to the absence of fine structure and texture information, existing fusion-based few-shot image generation methods suffer from unsatisfactory generation quality and diversity. To address this problem, we propose a novel feature Equalization fusion Generative Adversarial Network (EqGAN) for few-shot image generation. Unlike existing fusion strategies that rely on either deep features or local representations, we design two separate branches to fuse structures and textures by disentangling encoded features into shallow and deep contents. To refine image contents at all feature levels, we equalize the fused structure and texture semantics at different scales and supplement the decoder with richer information by skip connections. Since the fused structures and textures may be inconsistent with each other, we devise a consistent equalization loss between the equalized features and the intermediate output of the decoder to further align the semantics. Comprehensive experiments on three public datasets demonstrate that, EqGAN not only significantly improves generation performance with FID score (by up to 32.7%) and LPIPS score (by up to 4.19%), but also outperforms the state-of-the-arts in terms of accuracy (by up to 1.97%) for downstream classification tasks.