Geometric camera calibration is essential for applications that require understanding image perspective. We introduce Perspective Fields, a representation modeling local perspective properties using per-pixel information about the camera view, parameterized by an up vector and latitude value. This approach offers several benefits, such as minimal assumptions about the camera model, invariance or equivariance to common image editing operations, and improved interpretability aligned with human perception. We train a neural network to predict Perspective Fields, which can be easily converted to calibration parameters. Our method outperforms camera calibration-based approaches in various scenarios and is demonstrated in image compositing applications.

For each pixel location, the Perspective Field consists of a unit Up-vector and Latitude. The Up-vector is the projection of the up direction, shown in Green arrows. In perspective projection, it points to the vertical vanishing point. The Latitude of each pixel is defined as the angle between the incoming ray and the horizontal plane. We show it using contour line: -π/2 π/2. Note 0° is at the horizon.