How arthurdelannoyazerty/mesh-simplification-gnn works
Main System Components
a) Data Processing Layer
• 3D model input handling (OBJ/STL files)
• Conversion pipeline (convert_OBJ_to_STL.py)
b) Core Model Architecture
• Point Picker Module
• Triangle Generator Module
• GNN (Graph Neural Network) Components
c) Neural Network Layers
• Various specialized layers (organized hierarchically)
• Model training components
Component Relationships
[3D Model Input] → [Data Processing] → [Point Picker] → [Triangle Generator] → [Mesh Output]
With detailed sub-components:
Model Layers:
• GNN Components
• Convolution Layers
• Attention Mechanisms
• MLP Networks
Specific Implementation Details
• Blue: Data input/output components
• Green: Core processing modules
• Yellow: Neural network layers
• Red: Training components
Key Architectural Elements to Highlight
a) Model Architecture:
• Point Picker (model_point_picker.py)
• Triangle Generator (model_triangle_generator.py)
• Graph Neural Network components (gnn.py, devconv_gnn.py)
b) Layer Organization:
• Attention mechanisms (sparse_attention_edge_predictor_layer.py)
• Convolution operations (devcon.py)
• Graph operations (knn.py, knn_simple.py)
• Basic neural components (mlp.py, mlp_gnn.py)
c) Training Pipeline:
• Point picker training (train_point_picker.py)
• Triangle generator training (train_triangle_generator.py)
• Loss functions (loss_point_picker.py, loss_triangle_generator.py)
Visualization Components
• Visualization modules (viz_model_point_picker_training.py, viz_model_training.py)
• How they interact with the training process
Technical Details
Specific technical elements:
• Input formats (OBJ/STL)
• Layer types and connections
• Loss function integration
• Training flow direction
• Visualization feedback mechanisms