Differentiable Rendering

Differentiable rendering studies how to compute useful gradients through image formation so that scene geometry, materials, lighting, motion, or graphics can be fitted to visual targets. Common approaches include soft rasterization, Gaussian and other point-based splatting, differentiable ray tracing, and analytic coverage methods for meshes and vector shapes. These tools support tasks such as 3D reconstruction, novel-view synthesis, relighting, vectorization, and fitting physical models to sensor data, often evaluated on image-quality benchmarks and real-world captures. A central challenge is balancing physically accurate visibility and light transport with smooth, stable gradients and rendering that is fast and memory-efficient.