Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

Questions about adding THIN_PRISM_FISHEYE support to the Caspar backend and possible GLOMAP integration

Aperta
  • Qualcosa non va?
#4,371 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
25/100
Tipo di issue
Funzionalità
Chiarezza
Da chiarire
Stato di attività
Tranquilla
Stack tecnologico
cpp
Ambito
computer-vision

Direzione di ricerca

Start by reading the Caspar COLMAP integration PR #4018 and its discussion, then identify where camera-model-specific Caspar factors are implemented or generated. The issue names no files or tests and asks broad questions about both THIN_PRISM_FISHEYE support and GLOMAP integration; clarify which work is in scope before starting. Done would require an agreed scope and validation expectations.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Hi everyone,

I am opening this as a separate issue, as suggested in the Caspar COLMAP integration PR, to keep the original PR discussion focused. #4018 @tordnat

First of all, thank you for the work on integrating Caspar into COLMAP. This is very interesting to me because I work a lot with larger SfM / Gaussian Splatting reconstruction pipelines, where bundle adjustment and mapper runtime can become important bottlenecks.

My current workflow is usually:

COLMAP feature_extractor
COLMAP sequential_matcher
GLOMAP mapper

I have built both COLMAP and GLOMAP against a CUDA-enabled Ceres setup, and the GPU-accelerated parts in GLOMAP have made a very significant difference for my reconstructions. In practice, GLOMAP has been both fast and very stable for many of my datasets.

I am interested in two related questions:
1. Adding THIN_PRISM_FISHEYE support to the Caspar backend

A lot of my datasets come from fisheye workflows, and THIN_PRISM_FISHEYE is an important COLMAP camera model in my pipeline.

I would be interested in trying to add THIN_PRISM_FISHEYE support to the Caspar backend myself, but I would appreciate some guidance on what exactly would be required.

In particular, I would like to understand:
• where the current camera-model-specific Caspar factors are implemented or generated
• what needs to be added for a new COLMAP camera model
• whether fixed intrinsics and optimized intrinsics require separate factor variants
• whether partially optimized intrinsics, such as fixed principal point or fixed extra parameters, need special handling
• whether there are any known limitations for strongly distorted fisheye models in the current Caspar formulation
• what tests or benchmarks would be expected to validate such an implementation

My goal would be to first make this work in COLMAP and then benchmark it against my current GLOMAP + CUDA/Ceres workflow on real fisheye datasets.
2. Possible future Caspar integration into GLOMAP

The second question is more general and forward-looking.

Since GLOMAP is a very important part of my current reconstruction workflow, I would also like to understand what would be required, in principle, to make Caspar usable from GLOMAP as well.

I understand that this is probably not just a build flag or a drop-in replacement for Ceres, but rather a real integration effort. Still, it would be useful to understand the expected scope.

For example:
• Which parts of GLOMAP currently using Ceres would be candidates for a Caspar backend?
• Would this mostly affect bundle adjustment, or also other optimization steps in GLOMAP?
• Is the Caspar COLMAP integration structured in a way that could eventually be reused by GLOMAP, or would most of the integration have to be implemented separately?
• Are there architectural limitations that would make GLOMAP support unrealistic or very difficult?

The motivation is simple: I have been using GLOMAP very successfully for quite some time and consistently get clean and fast reconstructions. If Caspar can provide a significant additional speedup, it would be very tempting to explore whether GLOMAP could benefit from it as well.

I would be happy to test, benchmark, and potentially contribute THIN_PRISM_FISHEYE support if you can point me in the right direction.

Thank you!

Lingua principale
C++
Stelle
12.9k
Fork
2.2k
Merge medio
1g 2h
PR unite (30g)
78

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di colmap/colmap

Tutte le issue di colmap/colmap

Issue simili

Altre issue su C++

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.