Experiments with tools for difficult panoramas\batch work in PTGui

[Dmitry “Fotowizard”]

Hi everyone,

I would like to share an article about a personal research project and a set of experiments around extending PTGui workflows.

The practical question behind this work is fairly simple: how can difficult panorama projects become easier to inspect, understand, and prepare before the normal PTGui optimisation and stitching process. I am particularly interested in aerials and architectural panoramas with many frames/rows, strong parallax changes, repeated structures, glass and reflections, moving objects, water, and a mix of close foreground with distant background.

PTGui remains the centre of the workflow: it is the tool for optimisation, stitching, rendering, and final finishing. My experiments work alongside it, using the same `.pts` projects. The aim is not to create another stitcher, but to explore additional layers of analysis, visual feedback, and model-assisted project preparation.

The work started with geometry utilities and vertical-structure analysis. It then evolved into an overlay that visualises frame links, control points, verticals, and project structure directly in Panorama Editor. More recently, I have been experimenting with a separate environment for control-point analysis: current matchers such as LoMa and LightGlue-based combinations, semantic segmentation, depth masks, reflective-surface masks, geometric validation, and reproducible batch presets.

A few directions I find especially promising are:

- quickly seeing the structure and distribution of links and CPs in a large project;
- detecting verticals for buildings and other scenes with strong vertical structures;
- using segmentation to limit CP searches away from unstable classes such as sky, water, traffic, people, or vegetation;
- separating foreground and background with depth masks instead of asking one matching pass to explain both;
- using stronger modern correspondence models on difficult pairs, then filtering and reviewing the results before they are written to the `.pts` project;
- turning a successful configuration for one project into a repeatable batch operation for a series.

This is not an open-source project or a commercial product yet, and it is not ready for release. At the moment, it is a personal R&D effort: a place to test ideas, compare models, and see which workflows are actually useful on real panorama sets. Still, I thought the results, clips, and technical notes could be interesting to people working with challenging PTGui projects, and perhaps to the PTGui team as well.

Maybe I just need to find the time to make it releasable, haha.

I collected the current results, examples, and sources in this article:

https://www.panomind.pro/panomind-evolution/