Field notes — Petite Ceinture, January 2026
Paris, 12e arrondissement — 06.01.2026
→ English below
Pour ce rendu, j’ai effectué plusieurs itérations afin d’atteindre le résultat attendu au moment de la capture. J’ai observé des artefacts apparaissant dans les zones où des piétons traversaient la scène. 3 tests d’entraînement Gaussian avec différents paramètres et datasets.
—
For this render, I went through several iterations in order to reach the result I was expecting at the time of capture. I observed artifacts appearing in areas where pedestrians crossed the scene. 3 Gaussian training tests with different components and datasets.
1 — Interactive scene
Gaussian Splatting — navigable scene (SuperSplat viewer)
2 — Traveling “Pont 1”
Les artefacts se révèlent là où je croise des piétons
Artifacts are revealed where I cross into pedestrians
Traveling through the reconstructed environment — Unreal Engine
Training
Method: Gaussian Splatting (PostShot)
Images: 625
Splats: 2.0M
Iterations: 150k
Alignment: RealityCapture
Processing
Post-clean: none
Capture
Camera: Insta360 X4
Projection: equirectangular
Capture duration: 2:06 min (extract)
Location: Petite Ceinture, Paris 12e — outdoor / snow
3 — Training “Pont 1” — iteration 1
Gaussian Splatting training — 150k iterations | Images: 484 | Splats: 2.0M (PostShot)
4 — Training “Pont 1” — iteration 2
Gaussian Splatting training — 100k iterations | Images: 484 | Splats: 2.0M (PostShot)
5 — Training “Focus Pont” — iteration 3
Gaussian Splatting training — 150k iterations | Images: 625 | Splats: 2.0M (PostShot)
6 — Raw 360° capture “Focus Pont”
Source footage — Insta360 X4, equirectangular projection, 2:36 min
7 — Raw 360° capture “Bridge 1”
Source footage — Insta360 X4, equirectangular projection, 2:06 min
8 — Scene captures
A field iteration of the SAMPLE(S) project — documenting reconstruction instabilities on the Petite Ceinture.
Nicolas Mimault, Bagnolet / Paris · nmimault@gmail.com