AMD presented new research at the I3D Symposium titled "PEPS: Positional Encoding Projected Sampling", introducing a new method for positional encoding that can improve neural texture compression.
How Does Neural Texture Compression Work?
Neural Texture Compression works by training what are known as INRs, or “Implicit Neural Representations,” to learn coordinate-to-signal functions. By projecting texture coordinates into a higher-dimensional embedding and feeding this information to a multi-layer perceptron, it's possible to represent and compress textures significantly.
PEPS: Positional Encoding Projected Sampling
PEPS introduces a new method to improve the efficiency of this process by changing how positional encoding is used. Typically, positional encoding projects low-dimensional coordinates into a higher-dimensional sine/cosine vector. PEPS builds upon this by treating each sine/cosine projection as a point on a Lissajous curve, then sampling the encoder/grid at those projected points, thereby increasing the amount of information represented by the INR.
The compromise here is that this size reduction comes with a corresponding increase in computation cost. In AMD's testing on a 9070xt, generating a 1024×1024 three-channel texture increased from 4.32ms with the BI-grid baseline to 5.47ms with Grid-PEPS, while the further optimized Grid-PinkPEPS version reduced that to 4.86ms. This performance penalty can be explained by the extra computations and memory access required for the additional sampling introduced with PEPS.
Applications of PEPS Beyond Neural Texture Compression
Beyond neural texture compression, this new technique also has scope for use in SDFs (signed distance functions), which are used in 3D rendering. SDFs are notorious for requiring high-res grids that gobble up VRAM, so optimizing memory usage through lightweight neural compression models is important. When testing on the Pitted Stonefish SDF, Grid-PEPS was able to roughly match the IoU (Intersection Over Union, or how closely the reconstructed 3D shape overlaps the original) of non-PEPS methods with 8x more encoder parameters.
Implication For FSR Suite: Don't Expect To Run This On Your Radeon GPU Any Time Soon
While this research is certainly interesting from a technical standpoint, it's hard to tell when it will become relevant for consumers. Currently, only NVIDIA has any kind of publicly available toolkits/demos for Neural Texture Compression, and there isn't a single game out there with a full NTC implementation. On the AMD side, support is even more sparse - in fact, AMD hasn't even officially come up with a brand name for it yet; in all their research, they use generic terms to refer to the technology. Regardless, it's good to see progress in this domain, especially as the RAMpocalypse ensures we'll be seeing 8GB GPUs well into the latter half of this decade.
News Source: AMD GPUOpen
About the author: Rayan is an aspiring Computer Engineer, currently pursuing his undergraduate studies. He built his first computer in the pandemic, and has been hooked on the hobby ever since. He brings a unique blend of academic knowledge and technical know-how to his articles, which include everything from detailed instructional guides to performance comparisons in wccftech hardware section. When not stressing out over finals or writing articles, you can find him reading fantasy books or hitting the gym.
Follow Wccftech on Google to get more of our news coverage in your feeds.
