It allows for real-time decompression similar to block texture compression on GPUs.
NVIDIA has presented a new neural compression technique for material textures allowing for on-demand, real-time decompression with random access similar to block texture compression on GPUs, enabling compression on disk and memory.
The researchers unlock two more levels of detail using low bitrate compression, with better image quality than advanced image compression techniques, such as AVIF and JPEG XL, can offer.
The idea is to compress multiple material textures and their mipmap chains together and decompress them using a small neural network optimized for each material. The researchers use a custom training implementation to achieve practical compression speeds. They say its performance surpasses that of general frameworks, like PyTorch.
The paper called "Random-Access Neural Compression of Material Textures" is accepted for SIGGRAPH 2023 and is available here. Make sure to check it out and don't forget to join our 80 Level Talent platform and our Telegram channel, follow us on Instagram and Twitter, where we share breakdowns, the latest news, awesome artworks, and more.