You can deploy this model and start earning money today!

rasmuspjohansson/
KDS_spt_laz

semantic-segmentation

superpoint-transformer

apache-2.0

Views

No views yet

KDS SPT LAZ model

Pretrained Superpoint Transformer (SPT) checkpoint for semantic segmentation on the toy LAZ dataset (vox025toy_laz_dataset) from laz-superpoint_transformer.

Model file

File	Description
`vox025toy_laz_best.ckpt`	PyTorch Lightning checkpoint (~2.8 MB)

Training details

Field	Value
Experiment	`experiment=semantic/vox025toy_laz_dataset`
Training run	`2026-06-16_11-14-06`
Best epoch	909 (monitored on `val/miou`)
Best val/mIoU	~41.15

Companion dataset

Training tiles are in the HuggingFace dataset rasmuspjohansson/KDS_laz_dataset. Download with ./scripts/download_toy_laz_dataset.sh from the project repo.

Download

From a clone of laz-superpoint_transformer:

./scripts/download_model.sh

This places the checkpoint at checkpoints/vox025toy_laz_best.ckpt.

Usage

bash

python predict_many.py \
  --ckpt_path checkpoints/vox025toy_laz_best.ckpt \
  --inputlaz /path/to/tiles \
  --output_folder /path/to/results

Docker wrapper:

bash

./scripts/run_predict_docker.sh \
  --ckpt checkpoints/vox025toy_laz_best.ckpt \
  --input /path/to/tiles \
  --output /path/to/results

Republish (maintainers)

To sync a new best checkpoint back to this HuggingFace model repo:

./scripts/upload_model.sh

Requires write access and a HuggingFace token (HF_TOKEN or my_huggingface_token.txt at the orchestrator repo root, or hftoken_write.txt in the laz-superpoint_transformer repo root).

rasmuspjohansson/KDS_spt_laz