3 months ago
- Neural texture compression uses latent textures and an MLP decoder for real-time shading.
- A hypernetwork is trained to output both latent features and MLP weights, matching current methods' quality.
- The approach can infer multiple decoders or create decoders for super-resolution.