# yewzijian/regtr architecture

> Architecture diagram of the GitHub repository yewzijian/regtr, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-01). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/yewzijian/regtr
- Repository: https://github.com/yewzijian/regtr

## Overview

REGTR is a point-cloud registration system: it loads and transforms point-pair data, predicts point correspondences and confidence with a KPConv-plus-transformer network, and uses those predictions for registration and evaluation without a RANSAC stage. Training, benchmark evaluation, demo visualization, 3DMatch overlap preparation, and shared geometry utilities are represented. Several dataset, loss, and evaluation implementations were not excerpted; their nodes reflect documented capabilities and tree paths, while unverified connections are omitted.

Main entry points: [`src/models/regtr.py`](https://github.com/yewzijian/regtr/blob/main/src/models/regtr.py) (REGTR network), [`src/trainer.py`](https://github.com/yewzijian/regtr/blob/main/src/trainer.py) (Training runtime), [`src/demo.py`](https://github.com/yewzijian/regtr/blob/main/src/demo.py) (Registration demo), [`src/data_loaders/modelnet.py`](https://github.com/yewzijian/regtr/blob/main/src/data_loaders/modelnet.py) (ModelNet loader)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_data["Data preparation"]
  node_datasets[("3DMatch and ModelNet")]
  node_modelnet_loader["ModelNet loader<br/>[modelnet.py]"]
  node_match_loader["3DMatch loader<br/>[threedmatch.py]"]
  node_augment["Pair transforms"]
  node_match_transforms["Pair transforms<br/>[transforms.py]"]
  node_overlap_prep["Overlap preparation"]
end

subgraph group_model["Registration model"]
  node_kpconv["KPConv encoder<br/>[kpconv.py]"]
  node_kpconv_blocks["KPConv blocks<br/>[kpconv_blocks.py]"]
  node_kernel_points["Kernel points<br/>[kernel_points.py]"]
  node_transformer["Transformer layers<br/>[transformers.py]"]
  node_position_embedding["Position embeddings"]
  node_regtr["REGTR network<br/>[regtr.py]"]
  node_losses["Registration losses"]
end

subgraph group_runtime["Training and evaluation"]
  node_trainer["Training runtime<br/>[trainer.py]"]
  node_train_entry["Training entry<br/>[train.py]"]
  node_demo["Registration demo<br/>[demo.py]"]
  node_evaluation["3DMatch evaluation"]
end

subgraph group_geometry["Geometry utilities"]
  node_se3_torch["Torch SE&#40;3&#41;<br/>[se3_torch.py]"]
  node_se3_numpy["NumPy SE&#40;3&#41;<br/>[se3_numpy.py]"]
  node_lie_torch["Torch Lie groups"]
  node_lie_numpy["NumPy Lie groups"]
end

subgraph group_visual["Visualization"]
  node_visualization["Point-cloud visualization<br/>[viz.py]"]
  node_cv_visualization["Visualization backend"]
end

node_user(("Researcher"))

node_user -->|"starts training"| node_train_entry
node_train_entry -->|"runs"| node_trainer
node_trainer -->|"trains model"| node_regtr
node_user -->|"runs demo"| node_demo
node_demo -->|"loads model"| node_regtr
node_demo -->|"visualizes result"| node_visualization
node_user -->|"runs evaluation"| node_evaluation
node_datasets -->|"provides samples"| node_modelnet_loader
node_datasets -->|"provides samples"| node_match_loader
node_modelnet_loader -->|"applies transforms"| node_augment
node_match_loader -->|"applies transforms"| node_match_transforms
node_user -->|"precomputes overlap"| node_overlap_prep
node_regtr -->|"encodes clouds"| node_kpconv
node_kpconv -->|"uses blocks"| node_kpconv_blocks
node_kpconv_blocks -->|"uses kernels"| node_kernel_points
node_regtr -->|"attends features"| node_transformer
node_regtr -->|"encodes positions"| node_position_embedding
node_regtr -->|"computes losses"| node_losses
node_trainer -.->|"consumes batches"| node_modelnet_loader
node_trainer -.->|"consumes batches"| node_match_loader
node_regtr -->|"uses pose operations"| node_se3_torch
node_se3_torch -->|"uses Lie groups"| node_lie_torch
node_se3_numpy -->|"uses Lie groups"| node_lie_numpy
node_visualization -->|"renders objects"| node_cv_visualization

click node_modelnet_loader "https://github.com/yewzijian/regtr/blob/main/src/data_loaders/modelnet.py"
click node_match_loader "https://github.com/yewzijian/regtr/blob/main/src/data_loaders/threedmatch.py"
click node_augment "https://github.com/yewzijian/regtr/blob/main/src/data_loaders/modelnet_transforms.py"
click node_match_transforms "https://github.com/yewzijian/regtr/blob/main/src/data_loaders/transforms.py"
click node_overlap_prep "https://github.com/yewzijian/regtr/blob/main/src/data_processing/compute_overlap_3dmatch.py"
click node_kpconv "https://github.com/yewzijian/regtr/blob/main/src/models/backbone_kpconv/kpconv.py"
click node_kpconv_blocks "https://github.com/yewzijian/regtr/blob/main/src/models/backbone_kpconv/kpconv_blocks.py"
click node_kernel_points "https://github.com/yewzijian/regtr/blob/main/src/models/backbone_kpconv/kernels/kernel_points.py"
click node_transformer "https://github.com/yewzijian/regtr/blob/main/src/models/transformer/transformers.py"
click node_position_embedding "https://github.com/yewzijian/regtr/blob/main/src/models/transformer/position_embedding.py"
click node_regtr "https://github.com/yewzijian/regtr/blob/main/src/models/regtr.py"
click node_losses "https://github.com/yewzijian/regtr/tree/main/src/models/losses"
click node_trainer "https://github.com/yewzijian/regtr/blob/main/src/trainer.py"
click node_train_entry "https://github.com/yewzijian/regtr/blob/main/src/train.py"
click node_demo "https://github.com/yewzijian/regtr/blob/main/src/demo.py"
click node_evaluation "https://github.com/yewzijian/regtr/blob/main/src/evaluate_3dmatch.py"
click node_se3_torch "https://github.com/yewzijian/regtr/blob/main/src/utils/se3_torch.py"
click node_se3_numpy "https://github.com/yewzijian/regtr/blob/main/src/utils/se3_numpy.py"
click node_lie_torch "https://github.com/yewzijian/regtr/tree/main/src/cvhelpers/lie/torch"
click node_lie_numpy "https://github.com/yewzijian/regtr/tree/main/src/cvhelpers/lie/numpy"
click node_visualization "https://github.com/yewzijian/regtr/blob/main/src/utils/viz.py"
click node_cv_visualization "https://github.com/yewzijian/regtr/tree/main/src/cvhelpers/visualization"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_datasets,node_modelnet_loader,node_match_loader,node_augment,node_match_transforms,node_overlap_prep toneBlue
class node_kpconv,node_kpconv_blocks,node_kernel_points,node_transformer,node_position_embedding,node_regtr,node_losses toneAmber
class node_trainer,node_train_entry,node_demo,node_evaluation toneMint
class node_se3_torch,node_se3_numpy,node_lie_torch,node_lie_numpy toneRose
class node_visualization,node_cv_visualization,node_user toneIndigo
```

## Components

### Data preparation

- **3DMatch and ModelNet**
- [ModelNet loader](https://github.com/yewzijian/regtr/blob/main/src/data_loaders/modelnet.py): `src/data_loaders/modelnet.py`
- [3DMatch loader](https://github.com/yewzijian/regtr/blob/main/src/data_loaders/threedmatch.py): `src/data_loaders/threedmatch.py`
- [Pair transforms](https://github.com/yewzijian/regtr/blob/main/src/data_loaders/modelnet_transforms.py): `src/data_loaders/modelnet_transforms.py`
- [Pair transforms](https://github.com/yewzijian/regtr/blob/main/src/data_loaders/transforms.py): `src/data_loaders/transforms.py`
- [Overlap preparation](https://github.com/yewzijian/regtr/blob/main/src/data_processing/compute_overlap_3dmatch.py): `src/data_processing/compute_overlap_3dmatch.py`

### Registration model

- [KPConv encoder](https://github.com/yewzijian/regtr/blob/main/src/models/backbone_kpconv/kpconv.py): `src/models/backbone_kpconv/kpconv.py`
- [KPConv blocks](https://github.com/yewzijian/regtr/blob/main/src/models/backbone_kpconv/kpconv_blocks.py): `src/models/backbone_kpconv/kpconv_blocks.py`
- [Kernel points](https://github.com/yewzijian/regtr/blob/main/src/models/backbone_kpconv/kernels/kernel_points.py): `src/models/backbone_kpconv/kernels/kernel_points.py`
- [Transformer layers](https://github.com/yewzijian/regtr/blob/main/src/models/transformer/transformers.py): `src/models/transformer/transformers.py`
- [Position embeddings](https://github.com/yewzijian/regtr/blob/main/src/models/transformer/position_embedding.py): `src/models/transformer/position_embedding.py`
- [REGTR network](https://github.com/yewzijian/regtr/blob/main/src/models/regtr.py): `src/models/regtr.py`
- [Registration losses](https://github.com/yewzijian/regtr/tree/main/src/models/losses): `src/models/losses`

### Training and evaluation

- [Training runtime](https://github.com/yewzijian/regtr/blob/main/src/trainer.py): `src/trainer.py`
- [Training entry](https://github.com/yewzijian/regtr/blob/main/src/train.py): `src/train.py`
- [Registration demo](https://github.com/yewzijian/regtr/blob/main/src/demo.py): `src/demo.py`
- [3DMatch evaluation](https://github.com/yewzijian/regtr/blob/main/src/evaluate_3dmatch.py): `src/evaluate_3dmatch.py`

### Geometry utilities

- [Torch SE(3)](https://github.com/yewzijian/regtr/blob/main/src/utils/se3_torch.py): `src/utils/se3_torch.py`
- [NumPy SE(3)](https://github.com/yewzijian/regtr/blob/main/src/utils/se3_numpy.py): `src/utils/se3_numpy.py`
- [Torch Lie groups](https://github.com/yewzijian/regtr/tree/main/src/cvhelpers/lie/torch): `src/cvhelpers/lie/torch`
- [NumPy Lie groups](https://github.com/yewzijian/regtr/tree/main/src/cvhelpers/lie/numpy): `src/cvhelpers/lie/numpy`

### Visualization

- [Point-cloud visualization](https://github.com/yewzijian/regtr/blob/main/src/utils/viz.py): `src/utils/viz.py`
- [Visualization backend](https://github.com/yewzijian/regtr/tree/main/src/cvhelpers/visualization): `src/cvhelpers/visualization`

### Other components

- **Researcher**

## Connections

- Researcher → Training entry: starts training (evidence: `src/train.py`)
- Training entry → Training runtime: runs (evidence: `src/train.py`)
- Training runtime → REGTR network: trains model (evidence: `src/trainer.py`)
- Researcher → Registration demo: runs demo (evidence: `src/demo.py`)
- Registration demo → REGTR network: loads model (evidence: `src/demo.py`)
- Registration demo → Point-cloud visualization: visualizes result (evidence: `src/demo.py`)
- Researcher → 3DMatch evaluation: runs evaluation
- 3DMatch and ModelNet → ModelNet loader: provides samples (evidence: `src/data_loaders/modelnet.py`)
- 3DMatch and ModelNet → 3DMatch loader: provides samples
- ModelNet loader → Pair transforms: applies transforms (evidence: `src/data_loaders/modelnet.py`)
- 3DMatch loader → Pair transforms: applies transforms (evidence: `src/data_loaders/transforms.py`)
- Researcher → Overlap preparation: precomputes overlap
- REGTR network → KPConv encoder: encodes clouds (evidence: `src/models/regtr.py`)
- KPConv encoder → KPConv blocks: uses blocks (evidence: `src/models/backbone_kpconv/kpconv.py`)
- KPConv blocks → Kernel points: uses kernels (evidence: `src/models/backbone_kpconv/kpconv_blocks.py`)
- REGTR network → Transformer layers: attends features (evidence: `src/models/regtr.py`)
- REGTR network → Position embeddings: encodes positions (evidence: `src/models/regtr.py`)
- REGTR network → Registration losses: computes losses (evidence: `src/models/regtr.py`)
- Training runtime → ModelNet loader: consumes batches
- Training runtime → 3DMatch loader: consumes batches
- REGTR network → Torch SE(3): uses pose operations (evidence: `src/models/regtr.py`)
- Torch SE(3) → Torch Lie groups: uses Lie groups (evidence: `src/utils/se3_torch.py`)
- NumPy SE(3) → NumPy Lie groups: uses Lie groups (evidence: `src/utils/se3_numpy.py`)
- Point-cloud visualization → Visualization backend: renders objects (evidence: `src/utils/viz.py`)
