Abstract:Woven fabrics are widely used in applications of realistic rendering, where real-time capability is also essential. However, rendering realistic woven fabrics in real time is challenging due to their complex structure and optical appearance, which cause aliasing and noise without many samples. The core of this issue is a multi-scale representation of the fabric shading model, which allows for a fast range query. Some previous neural methods deal with the issue at the cost of training on each material, which limits their practicality. In this paper, we propose a lightweight neural network to represent different types of woven fabrics at different scales. Thanks to the regularity and repetitiveness of woven fabric patterns, our network can encode fabric patterns and parameters as a small latent vector, which is later interpreted by a small decoder, enabling the representation of different types of fabrics. By applying the pixel's footprint as input, our network achieves multi-scale representation. Moreover, our network is fast and occupies little storage because of its lightweight structure. As a result, our method achieves rendering and editing woven fabrics at nearly 60 frames per second on an RTX 3090, showing a quality close to the ground truth and being free from visible aliasing and noise.
What problem does this paper attempt to address?
### What problems does this paper attempt to solve?
This paper aims to address the challenges of real - time rendering of realistic woven fabrics. Specifically, it mainly solves the following key problems:
1. **The contradiction between real - time performance and high - quality rendering**:
- Woven fabrics have complex micro - structures and strong anisotropic appearances, which make it difficult to achieve both high realism and low time cost in real - time rendering simultaneously. Traditional methods either require a large number of samples to avoid aliasing and noise or are unable to achieve real - time rendering.
2. **Multi - scale representation**:
- The complex structure of woven fabrics results in their optical properties varying at different scales, and the traditional linear mipmap method cannot effectively handle this multi - scale characteristic. Therefore, a method that can efficiently handle multi - scale representation is required.
3. **Universality and flexibility**:
- Some existing neural network methods need to be trained separately for each material, which limits their practical application range. In addition, these methods have difficulties in supporting fabric parameter adjustment.
### Solutions proposed in the paper
To solve the above problems, the authors propose a lightweight neural network model for representing multiple types of woven fabrics at different scales. The core innovation points of this method include:
- **Lightweight neural network**: By designing a small neural network, different types of fabrics can be efficiently represented, and real - time rendering and editing can be achieved.
- **Utilization of regularity and repeatability**: By taking advantage of the regularity and repeatability of woven fabric patterns, fabric patterns and parameters are encoded into a small latent vector, and these latent vectors are interpreted by a decoder to represent different fabric types.
- **Multi - scale representation**: Multi - scale representation is achieved by inputting the footprint information of pixels, ensuring rendering quality and efficiency.
- **Fast inference and editability**: The network structure is lightweight and can render at a speed close to 60 frames per second on an RTX 3090 GPU, while supporting real - time material editing.
### Experimental results
The experimental results show that this method achieves a quality close to the real value on multiple typical types of woven fabrics without obvious aliasing and noise. Compared with existing methods, this method is not only superior in quality, but also occupies extremely little storage space (less than 5 MB) and does not need to be trained separately for each material.
Through these improvements, this paper successfully solves multiple key problems in real - time rendering of woven fabrics and provides practical technical solutions for fields such as virtual reality, video games, and digital humans.