RenderingMetrics<T>

Helpers & Utilities · AiDotNet.NeuralRadianceFields.Metrics

Provides image quality metrics for evaluating neural rendering methods like NeRF.

For Beginners

When a neural network generates an image (like a new view of a 3D scene), we need ways to measure how good that image is compared to a real photograph. These metrics give us numbers that tell us how similar two images are.

How It Works

These metrics are essential for evaluating the quality of rendered images from neural radiance fields, 3D Gaussian splatting, and other view synthesis methods.

Methods

Method Summary
double MAE(Tensor<T> a, Tensor<T> b) Computes Mean Absolute Error between two tensors.
double MSE(Tensor<T> a, Tensor<T> b) Computes Mean Squared Error between two tensors.
double PSNR(Tensor<T> predicted, Tensor<T> groundTruth, double maxValue) Computes Peak Signal-to-Noise Ratio (PSNR) between two images.
double SSIM(Tensor<T> predicted, Tensor<T> groundTruth, int windowSize, double k1, double k2, double maxValue) Computes Structural Similarity Index Measure (SSIM) between two images.
double SimplifiedLPIPS(Tensor<T> predicted, Tensor<T> groundTruth) Computes a simplified perceptual loss (L1 in feature space) as a proxy for LPIPS.