Datasets:
Industrial 3D Semantic Segmentation β Web Display Assets
Display assets for the dataset website of a synthetic industrial RGB-D dataset for 3D
semantic segmentation, generated with NVIDIA Isaac Sim. Files in this repository back
the static website and are meant to be fetched directly via resolve URLs; the full
datasets and model checkpoints live in separate repositories.
Quantitative benchmark results are not included here; they will be published together with the thesis.
Layout
| Path | Content |
|---|---|
metrics/meta.json |
model registry and class lists |
metrics/compare_manifest.json |
manifest for the interactive prediction viewer |
media/compare/{frame}/ |
per-frame viewer panels: rgb, gt, {model}_pred, {model}_err (WebP, locked camera) |
media/qualitative_kuka/ |
KUKA hold-out qualitative panels (input / GT / predictions / error maps) |
media/noise/ |
depth and normal renderings at the four corruption levels |
media/modalities/ |
example RGB / depth / normal / semantic-GT frame |
media/sampling/ |
camera-sampling illustration renders |
Direct access
https://huggingface.co/datasets/min99ian/industrial-semseg-web/resolve/main/metrics/compare_manifest.json
All files are served with CORS enabled and can be fetched from browser JavaScript.
- Downloads last month
- 401