US Patent Application 18298852. TABLE DICTIONARIES FOR COMPRESSING NEURAL GRAPHICS PRIMITIVES simplified abstract
Contents
TABLE DICTIONARIES FOR COMPRESSING NEURAL GRAPHICS PRIMITIVES
Organization Name
Inventor(s)
Alexander Georg Keller of Berlin (DE)
[[:Category:Thomas M�ller-h�hne of Rheinfelden (DE)|Thomas M�ller-h�hne of Rheinfelden (DE)]][[Category:Thomas M�ller-h�hne of Rheinfelden (DE)]]
Towaki Takikawa of Toronto (CA)
TABLE DICTIONARIES FOR COMPRESSING NEURAL GRAPHICS PRIMITIVES - A simplified explanation of the abstract
This abstract first appeared for US patent application 18298852 titled 'TABLE DICTIONARIES FOR COMPRESSING NEURAL GRAPHICS PRIMITIVES
Simplified Explanation
- The patent application is about improving the performance of neural networks in terms of training speed, memory usage, and accuracy. - This is achieved by using a compressed neural graphics primitive representation. - A neural graphics primitive is a mathematical function that involves at least one neural network and is used to represent computer graphics such as images, 3D shapes, light fields, etc. - Instead of directly inputting data into the neural network, inputs are encoded into a higher dimensional space using a function. - The input consists of coordinates that identify a point in a multi-dimensional space. - The point is quantized and used to access an indexing codebook and a features codebook. - The indexing codebook stores learned index offsets, while the features codebook stores learned feature vectors. - The learned feature vectors are then used as inputs to the neural network. - The innovation aims to improve the efficiency and effectiveness of neural networks in processing computer graphics data.
Original Abstract Submitted
Neural network performance is improved in terms of training speed, memory footprint, and/or accuracy by learning a compressed neural graphics primitive representation. A neural graphics primitive is a mathematical function involving at least one neural network, used to represent a computer graphic, where the graphic can be an image, a 3D shape, a light field, a signed distance function, a radiance field, 2D video, volumetric video, etc. Instead of being input directly to a neural network, inputs are effectively mapped (encoded) into a higher dimensional space via a function. The input comprises coordinates used to identify a point within a d-dimensional space. The point is quantized and a set of vertex coordinates corresponding to the point are used to access an indexing codebook and a features codebook that store learned index offsets and learned feature vectors, respectively. The learned feature vectors are then provided as inputs to the neural network.
