Detailed tutorials

Start with the progressive Guides paths when solving an application. This section retains the deeper tutorials, derivations, complete configuration references and reproducibility protocols to which those paths lead.

Use Compose a sphere-native model when inventing a new sphere-native method. The How PanorAi executes computer vision on spherical images provides the complementary view by execution strategy.

Compose a new sphere-native method

• Compose a sphere-native model

Project and reconstruct panoramic views

• Quick start
• Work with a spherical image object
• Projection foundations: sphere, rays, views, samplers, and blenders
• Cubemap, batch, and autograd

Process directly on the sphere

• Tutorial: image processing on the sphere
• Clean spherical masks with morphology

Extract and match visual features

• Tutorial: spherical features and matching

Estimate pose and radial range

• Tutorial: spherical Essential matrix and two-view triangulation
• Tutorial: dense stereo from two spherical images
• Tutorial: resize-free spherical monocular depth

Reconstruct and navigate

• Tutorial: from matched panoramas to a multiview reconstruction
• Tutorial: spherical visual SLAM

Localize and associate semantic regions

• Tutorial: pretrained spherical FCN and class-activation maps
• From spherical CAMs to spatial object hypotheses

CAM channels are directional learned evidence. They are not presented as instance masks or calibrated pixel probabilities; the association tutorial shows how registered views and tracks can turn broad regions into auditable spatial hypotheses.

Understand strategies and evidence

• How PanorAi executes computer vision on spherical images
• Tutorial: benchmark speed and geometric accuracy