The depth map

Every run ends with a depth from focus pass. It is not optional and it is not an extra mode: each frame's focus measure is taken as the frame is folded into the fusion, so the information is already there, and on a 25-frame 45-megapixel run the pass costs no measurable wall time.

What comes out is a dense, sub-frame depth map — the fractional frame index at which each pixel came into focus — plus a confidence map saying how much the pass trusted each pixel's own measurement.

Depth from focus covers the algorithm and its parameters. This page is about what you see and what it is for.

Focus depth

Depth → Focus depth in the viewer.

The depth map itself, shown on the working grid. The lookup table switches between Gray and Turbo; Turbo is easier to read as a surface, gray is easier to judge as a measurement.

Values are frame indices, interpolated between frames — a pixel that peaked between frames 7 and 8 reads 7.4, not 7. That sub-frame precision is what makes the stereo and 3D outputs smooth rather than terraced.

The map is saved either as an eight-bit gray PNG scaled from its own minimum to maximum (good for looking at), or as a sixteen-bit PNG on a fixed scale where 65535 is the last frame (good for measuring, and the same encoding as the command line's --depth-raw).

Confidence

Depth → Confidence in the viewer.

The peak-ratio confidence of each pixel's focus profile, normalized so its 90th percentile is

Concretely, it is the weight the smoothing solve gave that pixel's own depth measurement.

Read it as a map of where the depth is measured versus where it is inferred:

Bright — the pixel had a clear, unambiguous focus peak. The depth there is measured.
Dark — the pixel was flat, noisy, or had two competing peaks. Its depth was filled in from confident neighbors, without crossing image edges.

Large dark areas are a warning that the depth map there is a plausible interpolation rather than an observation. That matters if you are about to build a 3D model from it, or measure something.

Ctrl + G on a pixel logs its depth and confidence values.

In focus

Depth → In focus in the viewer. This is Helicon Focus's "source map" idea.

It shows the scrubbed frame's own pixels, in color, but darkened by how far the depth map puts each pixel from that frame:

Full brightness within ±0.5 frames
8 % beyond ±0.75 frames
Linear between

So the plane of focus stands out with a crisp edge, and scrubbing sweeps it through the scene. It is the most direct answer to "which parts of the result came from this frame?"

The dimmed, out-of-focus part keeps its outlines: the local contrast of its luminance is added back in gray, so blurred edges and fibers read as light lines against the dimmed color, and you can still see what you are looking at.

The page builds a preview from the proxy and the depth map's working grid immediately; the engine then returns the full-resolution aligned frame rendered the same way from the guided-upsampled depth map (about a second at 45 megapixels), which replaces the preview.

Compare it against LAP to see exactly where the fusion took its pixels from.

The depth slice

slice on the Focus depth and Confidence views.

Paints a 60 % magenta band over the pixels the depth map assigns to the scrubbed frame (rounded). Scrolling through the stack sweeps the band through the depth map. It is the quickest way to see whether the depth map is monotonic and sensible, and to spot a region that jumped.

Focus peaking

peaking on the Source view, with a threshold stepper.

The same magenta band, but on the source frames and computed from contrast rather than from the depth map. A pixel is painted for frame i when that frame's contrast is at least threshold × the largest contrast any frame in the stack has at that pixel — and when that maximum is above a small noise floor, so empty sky does not light up.

Scrubbing shows the in-focus band sweep through the scene, and each filmstrip entry shows its own percentage in focus.

This is the check to run before trusting a stack: if consecutive frames' bands do not overlap, the focus step was too large and there is depth nothing saw sharp. No setting will fix that.

The measure comes from the level-1 region energy the fusion already built, area-averaged to proxy resolution during the run, so it costs nothing extra.

What the depth map is good for

Judging the capture. Gaps, jumps and large low-confidence regions all tell you something about the shoot.
Rendering. The depth-map rendering blends frames by depth instead of by pyramid energy, which handles some subjects better than the pyramid does.
Relief. Stereo pairs, rocking animations and 3D models are all the result sheared or raised by this map.
Measurement. With the sixteen-bit fixed-scale export, the depth map is a real measurement of relative distance in units of frame index. Convert to physical units with your step size.

Accuracy

The browser's depth map matches the native one on the same frames to 4 × 10⁻⁴ frames — the limit of the sixteen-bit quantization the slices are stored in — and the confidence to 3 × 10⁻³ on average. About 0.1 % of pixels differ by more; those are near-ties between two peaks, where the two implementations fall opposite ways.