# Self-Supervised Contrastive Learning and GAN-Based Denoising for High-Fidelity HumanNeRF Images

## TL;DR

This paper introduces CD-GAN, a new method that improves the quality of human images generated by HumanNeRF by removing noise and enhancing details like skin texture and clothing wrinkles.

## Contribution

CD-GAN is a novel self-supervised denoising framework combining contrastive learning and GANs, eliminating the need for clean ground truth data.

## Key findings

- CD-GAN effectively removes complex, structured noise from HumanNeRF images without requiring paired clean data.
- The method preserves and enhances high-frequency details such as skin texture and clothing wrinkles.
- Experimental results show improved image quality and realism, supporting applications like virtual reality and digital avatars.

## Abstract

What are the main findings?
This paper proposes CD-GAN, a novel self-supervised denoising framework that effectively combines contrastive learning with Generative Adversarial Networks (GANs) to remove complex, structured noise from images generated by HumanNeRF.The method operates without the need for any paired “clean” ground truth data by leveraging the intrinsic stochasticity of the HumanNeRF rendering process to construct positive and negative sample pairs for training.

This paper proposes CD-GAN, a novel self-supervised denoising framework that effectively combines contrastive learning with Generative Adversarial Networks (GANs) to remove complex, structured noise from images generated by HumanNeRF.

The method operates without the need for any paired “clean” ground truth data by leveraging the intrinsic stochasticity of the HumanNeRF rendering process to construct positive and negative sample pairs for training.

What are the implications of the main findings?
The proposed method significantly enhances the visual quality of dynamic human neural renderings by not only suppressing noise but also preserving and enhancing critical high-frequency details such as skin texture and clothing wrinkles, providing crucial support for downstream applications like virtual reality and digital avatars.The innovative integration of contrastive learning within a self-supervised denoising paradigm offers a new and extensible solution for addressing image quality issues in other neural rendering scenarios.

The proposed method significantly enhances the visual quality of dynamic human neural renderings by not only suppressing noise but also preserving and enhancing critical high-frequency details such as skin texture and clothing wrinkles, providing crucial support for downstream applications like virtual reality and digital avatars.

The innovative integration of contrastive learning within a self-supervised denoising paradigm offers a new and extensible solution for addressing image quality issues in other neural rendering scenarios.

To address the prevalent noise issue in images generated by HumanNeRF, this paper proposes an image denoising method that combines self-supervised contrastive learning and Generative Adversarial Networks (GANs). While HumanNeRF excels in realistic 3D human reconstruction tasks, its generated images often suffer from noise and detail loss due to incomplete training data and sampling noise during the rendering process. To solve this problem, our method first utilizes a self-supervised contrastive learning strategy to construct positive and negative sample pairs, enabling the network to effectively distinguish between noise and human detail features without external labels. Secondly, it introduces a Generative Adversarial Network, where the adversarial training between the generator and discriminator further enhances the detail representation and overall realism of the images. Experimental results demonstrate that the proposed method can effectively remove noise from HumanNeRF images while significantly improving detail fidelity, ultimately yielding higher-quality human images and providing crucial support for subsequent 3D human reconstruction and realistic rendering.