How canaltinigne/deepheightweight works
Project Type & Purpose
• A deep-learning pipeline for estimating human height and weight from images.
• Implements two parallel networks (height vs. weight) with shared design pattern.
Key Components (from file tree & README)
• Data Layer
– dataset.py / dataset_weight.py: data loading, augmentation, batching
– dataset_test.py / dataset_weight_test.py: unit tests for data pipelines
• Model Layer
– network.py / network_weight.py: PyTorch model definitions
– loss.py: custom loss functions
• Training Orchestration
– train.py / train_weight.py: training loops, optimizer setup, checkpoint saving
– logs/: training logs (e.g., TensorBoard)
• Artifacts & Analysis
– models/: serialized checkpoints (2 GB each)
– Results_HW.ipynb & images/: evaluation, visual results
• Environment & Configuration
– environment.yml: conda environment spec
Component Relationships & Data Flow
Data Ingestion → Model Definition → Training Loop →
Logs & Checkpoints → Evaluation Notebook
(Two parallel branches for height and weight networks, sharing the same flow)
Architectural Patterns & Principles
• Modular separation: data / model / training / evaluation
• Reproducibility: environment.yml, unit tests, fixed directory layout
• Experiment tracking: logs/ + models/ persistence