How meidachen/stpls3d works
STPLS3D is an annotated aerial point-cloud dataset accompanied by example semantic- and instance-segmentation benchmarks. The user-facing workflow is to prepare/load point clouds, train a segmentation model, then evaluate predictions. The repository contains several largely independent model implementations rather than one shared application pipeline. Edges below are limited to relationships visible in the source index; documented training and evaluation capabilities are represented, but some links to unsampled entry points remain unverified.
Main entry points: STPLS3D.py KPConv loader · train_STPLS3D.py KPConv training · hais.py HAIS model · architectures.py KPConv architecture
Components 23
Dataset handling
- STPLS3D data
- KPConv loader
KPConv-PyTorch/datasets/STPLS3D.py - Cloud utilities
KPConv-PyTorch/datasets/common.py
Semantic benchmarks
- KPConv training
KPConv-PyTorch/train_STPLS3D.py - KPConv architecture
KPConv-PyTorch/models/architectures.py - KPConv blocks
KPConv-PyTorch/models/blocks.py - Kernel points
KPConv-PyTorch/kernels/kernel_points.py - KPConv evaluation
KPConv-PyTorch/utils/tester.py - Segmentation metrics
KPConv-PyTorch/utils/metrics.py - RandLA-Net
RandLA-Net/RandLANet.py - RandLA workflow
RandLA-Net/main_STPLS3D.py - SCF-Net
SCF-Net/SCFNet.py - SCF workflow
SCF-Net/STPLS3D_main.py - Point Transformer training
point-transformer/tool/train_stpls.py - Point Transformer model
point-transformer/model/pointtransformer/pointtransformer_seg.py - Point transforms
point-transformer/util/transform.py
Instance benchmark
- HAIS training
HAIS/train.py - Instance data loader
HAIS/data/stpls3d_inst.py - HAIS model
HAIS/model/hais/hais.py - Challenge evaluation
HAIS/STPLS3DInstanceSegmentationChallenge_Codalab_Evaluate.py
Point operations
- HAIS point ops
HAIS/lib/hais_ops/functions/hais_ops.py - Sparse modules
HAIS/lib/spconv/spconv/modules.py
Other components
- Researcher
Connections 21
- Researcher → STPLS3D data: uses data
- STPLS3D data → KPConv loader: loads clouds
- KPConv training → KPConv loader: trains on
- KPConv training → KPConv architecture: trains model
- KPConv architecture → KPConv blocks: uses blocks
KPConv-PyTorch/models/blocks.py - KPConv blocks → Kernel points: uses kernels
KPConv-PyTorch/models/blocks.py - KPConv loader → Cloud utilities: uses utilities
KPConv-PyTorch/datasets/STPLS3D.py - KPConv evaluation → Segmentation metrics: computes metrics
KPConv-PyTorch/utils/tester.py - Researcher → RandLA workflow: runs benchmark
- RandLA workflow → RandLA-Net: trains model
- Researcher → SCF workflow: runs benchmark
- SCF workflow → SCF-Net: trains model
- Researcher → Point Transformer training: runs training
- Point Transformer training → Point Transformer model: trains model
point-transformer/tool/train_stpls.py - Point Transformer model → Point transforms: uses transforms
- Researcher → HAIS training: runs training
- HAIS training → Instance data loader: loads instances
- HAIS model → HAIS point ops: calls operations
HAIS/model/hais/hais.py - HAIS model → Sparse modules: uses modules
HAIS/model/hais/hais.py - Instance data loader → HAIS point ops: uses operations
HAIS/data/stpls3d_inst.py - Researcher → Challenge evaluation: submits results