# eth-ait/gaussianhaircut architecture

> Architecture diagram of the GitHub repository eth-ait/gaussianhaircut, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2024-12-28). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/eth-ait/gaussianhaircut
- Repository: https://github.com/eth-ait/gaussianhaircut

## Overview

##### Main Components

a) Input Processing Pipeline:
- Video Input Component
- Preprocessing Module (from src/preprocessing/)
- Image Processing
- Face Alignment
- Mask Generation
- FLAME Fitting Integration

b) Core Processing Systems:
- Gaussian Model System
- Base Gaussian Model
- Strand-Aligned Gaussian Model
- Latent Strand Model
- CUDA Rasterization Engine (from ext/diff_gaussian_rasterization_hair/)
- Training Pipeline

c) Rendering System:
- Gaussian Renderer
- Post-processing Pipeline
- Blender Integration

##### Component Relationships

- Data flow from raw video input through preprocessing
- Training pipeline connecting Gaussian models with the rasterization engine
- Rendering pipeline showing both real-time visualization and final output generation

##### Specific Implementation Details

**[Input Video] → [Preprocessing Pipeline]**
- ↓
**[Training System] ← → [CUDA Rasterization Engine]**
- ↓
**[Gaussian Models] → [Rendering Pipeline] → [Output Visualization]**

##### Important Architectural Elements to Highlight

a) External Dependencies:
- CUDA 11.8 integration
- Blender 3.6 integration
- Neural Haircut components
- FLAME fitting pipeline

b) Core Processing Pipeline:
- Strand-aligned 3D Gaussian processing
- Differential rasterization system
- Training loop for Gaussian models

##### Technical Details

- CUDA-accelerated components
- Model architecture showing:
- Gaussian model variants (base, strands, latent)
- Rasterization pipeline
- Training systems

## Diagram (Mermaid)

```mermaid
graph TB
    subgraph Input
        VI[Video Input]
        PP[Preprocessing Pipeline]:::preproc
        click PP "https://github.com/eth-ait/GaussianHaircut/tree/main/src/preprocessing/"
        PP_FA[Face Alignment]:::preproc
        click PP_FA "https://github.com/eth-ait/GaussianHaircut/blob/main/src/preprocessing/calc_face_alignment.py"
        PP_MG[Mask Generation]:::preproc
        click PP_MG "https://github.com/eth-ait/GaussianHaircut/blob/main/src/preprocessing/calc_masks.py"
        PP_FI[FLAME Integration]:::preproc
        click PP_FI "https://github.com/eth-ait/GaussianHaircut/blob/main/src/preprocessing/filter_flame_intersections.py"
        
        VI --> PP
        PP --> PP_FA
        PP --> PP_MG
        PP --> PP_FI
    end

    subgraph "Core Processing"
        GM[Base Gaussian Model]:::model
        click GM "https://github.com/eth-ait/GaussianHaircut/blob/main/src/scene/gaussian_model.py"
        GMS[Strand-Aligned Model]:::model
        click GMS "https://github.com/eth-ait/GaussianHaircut/blob/main/src/scene/gaussian_model_strands.py"
        GML[Latent Strand Model]:::model
        click GML "https://github.com/eth-ait/GaussianHaircut/blob/main/src/scene/gaussian_model_latent_strands.py"
        
        CUDA[CUDA Rasterization Engine]:::cuda
        click CUDA "https://github.com/eth-ait/GaussianHaircut/tree/main/ext/diff_gaussian_rasterization_hair/"
        CUDA_F[Forward Pass]:::cuda
        click CUDA_F "https://github.com/eth-ait/GaussianHaircut/blob/main/ext/diff_gaussian_rasterization_hair/cuda_rasterizer/forward.cu"
        CUDA_B[Backward Pass]:::cuda
        click CUDA_B "https://github.com/eth-ait/GaussianHaircut/blob/main/ext/diff_gaussian_rasterization_hair/cuda_rasterizer/backward.cu"
        CUDA_R[Rasterizer Core]:::cuda
        click CUDA_R "https://github.com/eth-ait/GaussianHaircut/blob/main/ext/diff_gaussian_rasterization_hair/cuda_rasterizer/rasterizer_impl.cu"
        
        CUDA --> CUDA_F
        CUDA --> CUDA_B
        CUDA --> CUDA_R
    end

    subgraph Training
        TR_B[Base Training]:::training
        click TR_B "https://github.com/eth-ait/GaussianHaircut/blob/main/src/train_gaussians.py"
        TR_S[Strands Training]:::training
        click TR_S "https://github.com/eth-ait/GaussianHaircut/blob/main/src/train_strands.py"
        TR_L[Latent Strands Training]:::training
        click TR_L "https://github.com/eth-ait/GaussianHaircut/blob/main/src/train_latent_strands.py"
    end

    subgraph "Rendering & Output"
        GR[Gaussian Renderer]:::render
        click GR "https://github.com/eth-ait/GaussianHaircut/tree/main/src/gaussian_renderer/"
        PP_P[Post-processing Pipeline]:::render
        click PP_P "https://github.com/eth-ait/GaussianHaircut/tree/main/src/postprocessing/"
        BI[Blender Integration]:::render
        click BI "https://github.com/eth-ait/GaussianHaircut/blob/main/src/postprocessing/main.blend"
        
        GR --> PP_P
        PP_P --> BI
    end

    subgraph Utilities
        CU[Camera Utils]:::util
        click CU "https://github.com/eth-ait/GaussianHaircut/blob/main/src/utils/camera_utils.py"
        LU[Loss Utils]:::util
        click LU "https://github.com/eth-ait/GaussianHaircut/blob/main/src/utils/loss_utils.py"
        GU[Graphics Utils]:::util
        click GU "https://github.com/eth-ait/GaussianHaircut/blob/main/src/utils/graphics_utils.py"
    end

    subgraph Evaluation
        ME[Metrics]:::eval
        click ME "https://github.com/eth-ait/GaussianHaircut/blob/main/src/metrics.py"
        FE[Full Evaluation]:::eval
        click FE "https://github.com/eth-ait/GaussianHaircut/blob/main/src/full_eval.py"
    end

    %% Connections between components
    PP_FA & PP_MG & PP_FI --> GM
    GM --> GMS --> GML
    GM & GMS & GML <--> CUDA
    TR_B --> GM
    TR_S --> GMS
    TR_L --> GML
    GM & GMS & GML --> GR
    CU & LU & GU --> TR_B & TR_S & TR_L
    GR --> ME --> FE

    %% Styling
    classDef preproc fill:#a8e6cf
    classDef model fill:#dcedc1
    classDef cuda fill:#ffd3b6
    classDef training fill:#ffaaa5
    classDef render fill:#ff8b94
    classDef util fill:#98ddca
    classDef eval fill:#d4a5a5
```
