Stogram: Stereogram Viewer

English | Русский

An interactive autostereogram and depth map viewer. It reconstructs depth and renders it in Human Eye mode, as discrete layers, or as a shaded 3D mesh. Geometry and texture can be replaced independently.

Autostereogram

↓
Depth map

↓
3D Mesh or Human Eye view

Features

• Depth map reconstruction from a color autostereogram.
• Loading grayscale images as ready-made depth maps.
• Human Eye, Layers, 3D Mesh, and 2D Depth Map modes.
• Convex and concave geometry.
• Independent texture replacement without losing the original stereogram and depth.
• WebGPU-accelerated reconstruction and filtering with a CPU fallback.
• Export of the depth map as PNG and the closed surface mesh as binary STL.

Loading Images

A color image is treated as a stereogram and also becomes the texture.

A grayscale image is treated as a ready-made depth map.

A depth map replaces the geometry while preserving the current texture. If there is no texture yet, the map itself is used.

Images can also be supplied through URL parameters :

• depth=dolphins.jpg
• texture=camel.jpg

Example.

Interface

Control	Action
＋	Load a stereogram or depth map
▧	Replace the texture
↻	Recalculate the depth map
⊕	Reset the camera
▶	Auto-rotation
❍	3D Mesh
≡	Layers
◉	Human Eye
◩	2D depth map
◠ / ◡	Convex / concave geometry
⛶	Fullscreen

Slider	Action
↕︎	Depth scale
↔	Stereogram period

Controls

• Drag — rotation.
• Mouse wheel or pinch — zoom.
• Double-click — reset the camera.
• Tap the texture preview — fullscreen stereogram viewing.
• Regular drop — load a geometry source.
• Drop onto ▧ or Shift + drop — replace the texture.

Depth Map Calculation

Stereogram
    ↓ resize to a maximum of 512 px + grayscale
Global period search
    ↓ refinement using local blocks
5×5 Census
    ↓ cost volume for possible disparities
Four-direction SGM (WebGPU when available)
    ↓ best disparity selection + confidence
3 local ICM refinement passes
    ↓ normalization and background detection
Island and MAD spike removal
    ↓ hole filling and contour smoothing
Weighted median
    ↓ mask-aware Gaussian smoothing
Soft contour coverage (antialiasing)
    ↓ robust 1–99% depth normalization
Ready depth map
    ↓
Eye View, discrete layers, smoothed closed 3D mesh, PNG or STL