ImageView<T>

Models & Types · AiDotNet.NeuralRadianceFields.Data

A single photo + camera pose used as input to image-space (photometric) radiance-field training (#1834). Reference NeRF / GS implementations require the caller to hand-assemble ray samples from photos; this type is the canonical shape a radiance-field loader emits so the model receives everything it needs to sample rays and compute a photometric loss.

How It Works

Reference impls fragment this data across a config file (poses), a directory of image files (photos), and hard-coded assumptions (focal length must be passed by hand). ImageView unifies them into one strongly-typed record with sensible nullable knobs (FocalLength is nullable — a null value asks the loader / facade to auto-detect from EXIF or fall back to a paper default derived from image size).

Constructors

Constructor Summary
ImageView(Tensor<T> photo, Vector<T> cameraPosition, Matrix<T> cameraRotation, T focalLength, ExifIntrinsics exif, LearnedPrior<T> prior)

Properties

Property Summary
ExifIntrinsics Exif { get; set; } Optional EXIF metadata bundle (35mm-equivalent focal length in mm + sensor width).
LearnedPrior<T> Prior { get; set; } Optional single-image reconstruction prior (Zero123 / TripoSR-style).
Matrix<T> CameraRotation { get; set; } Camera-to-world rotation matrix [3, 3].
T FocalLength { get; set; } Focal length in pixels.
Tensor<T> Photo { get; set; } Photo tensor of shape [H, W, 3] (RGB) or [H, W, 4] (RGBA — alpha ignored by the model, preserved for compositing).
Vector<T> CameraPosition { get; set; } Camera origin in world coordinates [3].
int Height { get; } Image height (pixels).
int Width { get; } Image width (pixels).

Methods

Method Summary
double ResolveFocalLengthInPixels() Resolves the effective focal length in pixels using the precedence order the loader + facade rely on: (1) explicit FocalLength the caller passed, (2) EXIF-derived focal computed from Exif and the photo’s pixel dimensions, (3) nerfstudio…