scatter_plot KOS Classify

Deep-learning point-cloud classification with a live 3D viewer, a 49-tool routine toolbox (ground, vegetation, buildings, noise…), fence & cross-section editing, macro pipelines, and on-box or local-GPU model training.

5Models
5Clouds
4Train runs

queued0%
Engine console
Fine classes
Macro groups
scatter_plot completed
classified.las
5,677,589 pts · 7/31/2026
scatter_plot completed
edge.pts
2,000 pts · 7/31/2026
scatter_plot completed
edge.xyz
2,000 pts · 7/31/2026
scatter_plot completed
edge_hdr.pts
1,500 pts · 7/31/2026
scatter_plot completed
test_import.laz
5,000 pts · 7/31/2026

Train a new model

Teacher–student training on procedurally generated synthetic scenes (no dataset needed). Runs on the server — larger runs take longer.

dataset Train on your labelled data

Upload classified clouds (.las/.laz/.ply/.pcd/.xyz) that already carry ground-truth class codes. They're split into Train / Validation / Test — 70 / 15 / 15, or 80 / 10 / 10 once you upload 20+ files — and used to train a real model (is_real_data). Runs on the server (CPU); for a GPU, use the Desktop app and pick this dataset folder.

training0%

Training runs

model_training
model_1785430750993.pth
completed 20 epochs · scenes 7/30/2026, 4:59:10 PM
model_training
model_1785430677446.pth
completed 20 epochs · scenes 7/30/2026, 4:57:57 PM
model_training
model_1785349769742.pth
completed 1 epochs · 3 scenes 7/29/2026, 6:29:29 PM
model_training
model_1784196741270.pth
completed 12 epochs · 6 scenes 7/16/2026, 10:12:21 AM
deployed_code kos_pointnet_v4 Active KOS Classify
Architecture: PointNet
Trained on: synthetic_bootstrap (synthetic)
Validation: 93.5% accuracy · 0.609 mIoU
18 epochs · 8 scenes
6.5 MB · 8/6/2026
deployed_code Built-in model Legacy
Architecture: RegionCNN
Trained on: synthetic_bootstrap (synthetic)
261 KB · 8/6/2026
download
deployed_code model_1785430750993 KOS Classify
Architecture: PointNet++
Trained on: ds_1785430672104_4eb96b
Validation: 87.8% accuracy · 0.794 mIoU
20 epochs · scenes
1.1 MB · 7/30/2026
download
deployed_code model_1785430677446 KOS Classify
Architecture: PointNet++
Trained on: ds_1785430672104_4eb96b
Validation: 87.8% accuracy · 0.794 mIoU
20 epochs · scenes
1.1 MB · 7/30/2026
download
deployed_code model_1785349769742 KOS Classify
Architecture: RandLA-Net
Trained on: synthetic_bootstrap (synthetic)
Validation: 57.3% accuracy · 0.096 mIoU
1 epochs · 3 scenes
2.7 MB · 7/29/2026
download

memory Local-GPU desktop & CLI

Train and classify on your own GPU, then push results here. Heavy models (RandLA-Net and the CUDA-only ones) run far faster locally. Download the app or CLI, sign in with a token, and pick your GPU.

Weights leave the server only through the app, are used in a temp folder, and securely wiped after each run. Results and any trained model are pushed straight back here.

key Access tokens

Paste one into the app/CLI to sign in. A token is shown only once — copy it immediately.