Depth Learning Method of Many Light Rendering based on Matrix Row and Column Sampling

摘要

In this paper, a deep learning method for multi light source rendering based on matrix row and column sampling is proposed, which includes the following steps: Step 1, establish a light matrix according to the scene, in which each column represents all sampling points illuminated by one light source, and each row represents one sampling point illuminated by all light sources; step 2, randomly select several rows from the multi light source matrix to form a shrinkage Subtraction matrix; step 3, drawing quadratic random reduction matrix and primary random reduction matrix respectively for different viewpoints; step 4, training a depth neural network by using the image pairs drawn by primary reduction matrix and quadratic reduction matrix. Step 5, using the trained depth neural network, in the real-time high reality rendering, first draw the matrix image obtained by the second reduction, then input the image to the depth neural network, the output image of the depth neural network is the complete rendering image we want to get. The invention transforms the multi light source rendering problem under the complex scene into the training and learning problem of the depth neural network, obtains the better rendering image through the processing of the depth neural network, and improves the rendering efficiency and real-time performance. It can be applied to scenes with real-time and high quality requirements.