# sanjayimmanuel24/lulc-efficientnet-ensemble architecture

> Architecture diagram of the GitHub repository sanjayimmanuel24/lulc-efficientnet-ensemble, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-05). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/sanjayimmanuel24/lulc-efficientnet-ensemble
- Repository: https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble

## Overview

This repository is an experimental land-cover classification system for multispectral Sentinel-2/EuroSAT imagery. The main workflow builds reproducible folds, reads and transforms tiles, trains a dual-branch classifier, and produces fused predictions and evaluation results; separate scripts support single-tile prediction, calibration, robustness experiments, and interpretation. The graph retains the README’s documented selective-prediction and learned-gate capabilities. Some script-to-module links are based on the SOURCE INDEX import references; unsampled implementation details are not inferred.

Main entry points: [`lulc-project/scripts/run_cross_validation.py`](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/run_cross_validation.py) (Cross-validation runner), [`lulc-project/src/models/ensemble.py`](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/ensemble.py) (Dual-branch model), [`lulc-project/scripts/predict.py`](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/predict.py) (Tile prediction), [`lulc-project/src/data/dataset.py`](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/dataset.py) (Tile dataset)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_data["Data preparation"]
  node_splits["Stratified splits<br/>[splits.py]"]
  node_dataset["Tile dataset<br/>[dataset.py]"]
  node_indices["Spectral indices"]
  node_transforms["Image transforms<br/>[transforms.py]"]
end

subgraph group_model["Classification model"]
  node_ensemble["Dual-branch model<br/>[ensemble.py]"]
  node_attention["ECA attention<br/>[attention.py]"]
  node_torch_fusion["Torch fusion<br/>[fusion_torch.py]"]
  node_fusion_reference["Fusion reference<br/>[fusion.py]"]
  node_baseline["ResNet baseline<br/>[baseline.py]"]
  node_loss["Smoothed loss"]
end

subgraph group_experiments["Training and prediction"]
  node_config["Run configuration<br/>[config.py]"]
  node_train_loop["Training loop<br/>[train.py]"]
  node_cv_runner["Cross-validation runner"]
  node_predict["Tile prediction<br/>[predict.py]"]
end

subgraph group_analysis["Evaluation and reliability"]
  node_calibration["Temperature scaling"]
  node_metrics["Classification metrics<br/>[metrics.py]"]
  node_statistics["Statistical tests"]
  node_compare["Variant comparison"]
  node_selective["Selective prediction"]
  node_gate["Learned fusion gate"]
  node_gradcam["Grad-CAM<br/>[gradcam.py]"]
  node_gradcam_script["Grad-CAM analysis"]
end

node_user(("Researcher"))
node_euro_dataset[("EuroSAT-MS tiles")]

node_user -->|"runs folds"| node_cv_runner
node_user -->|"requests predictions"| node_predict
node_euro_dataset -->|"reads tiles"| node_dataset
node_dataset -->|"computes indices"| node_indices
node_cv_runner -->|"loads settings"| node_config
node_cv_runner -->|"builds folds"| node_splits
node_cv_runner -->|"loads tiles"| node_dataset
node_cv_runner -->|"transforms inputs"| node_transforms
node_cv_runner -->|"trains model"| node_ensemble
node_cv_runner -->|"selects baseline"| node_baseline
node_cv_runner -->|"runs training"| node_train_loop
node_cv_runner -->|"fits calibration"| node_calibration
node_cv_runner -->|"computes metrics"| node_metrics
node_train_loop -->|"builds criterion"| node_loss
node_ensemble -->|"uses attention"| node_attention
node_ensemble -->|"uses fusion"| node_torch_fusion
node_predict -->|"loads tiles"| node_dataset
node_predict -->|"transforms inputs"| node_transforms
node_predict -->|"predicts classes"| node_ensemble
node_predict -->|"reports fusion"| node_fusion_reference
node_compare -->|"tests variants"| node_statistics
node_selective -->|"tests risk"| node_statistics
node_gate -->|"loads folds"| node_splits
node_gate -->|"loads tiles"| node_dataset
node_gate -->|"caches logits"| node_ensemble
node_gate -->|"compares fusion"| node_fusion_reference
node_gradcam_script -->|"generates maps"| node_gradcam
node_gradcam_script -->|"explains model"| node_ensemble

click node_config "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/config.py"
click node_splits "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/splits.py"
click node_dataset "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/dataset.py"
click node_indices "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/spectral_indices.py"
click node_transforms "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/transforms.py"
click node_ensemble "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/ensemble.py"
click node_attention "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/attention.py"
click node_torch_fusion "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/fusion_torch.py"
click node_fusion_reference "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/fusion.py"
click node_baseline "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/baseline.py"
click node_loss "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/losses/label_smoothing_torch.py"
click node_train_loop "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/training/train.py"
click node_cv_runner "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/run_cross_validation.py"
click node_predict "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/predict.py"
click node_calibration "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/calibration/temperature_scaling.py"
click node_metrics "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/evaluation/metrics.py"
click node_statistics "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/evaluation/statistical_tests.py"
click node_compare "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/compare_variants.py"
click node_selective "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/selective_prediction.py"
click node_gate "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/train_fusion_gate.py"
click node_gradcam "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/explainability/gradcam.py"
click node_gradcam_script "https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/make_gradcam_figures.py"

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_splits,node_dataset,node_indices,node_transforms toneBlue
class node_ensemble,node_attention,node_torch_fusion,node_fusion_reference,node_baseline,node_loss,node_euro_dataset toneAmber
class node_config,node_train_loop,node_cv_runner,node_predict toneMint
class node_calibration,node_metrics,node_statistics,node_compare,node_selective,node_gate,node_gradcam,node_gradcam_script toneRose
class node_user toneIndigo
```

## Components

### Data preparation

- [Stratified splits](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/splits.py): `lulc-project/src/data/splits.py`
- [Tile dataset](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/dataset.py): `lulc-project/src/data/dataset.py`
- [Spectral indices](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/spectral_indices.py): `lulc-project/src/data/spectral_indices.py`
- [Image transforms](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/data/transforms.py): `lulc-project/src/data/transforms.py`

### Classification model

- [Dual-branch model](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/ensemble.py): `lulc-project/src/models/ensemble.py`
- [ECA attention](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/attention.py): `lulc-project/src/models/attention.py`
- [Torch fusion](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/fusion_torch.py): `lulc-project/src/models/fusion_torch.py`
- [Fusion reference](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/fusion.py): `lulc-project/src/models/fusion.py`
- [ResNet baseline](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/models/baseline.py): `lulc-project/src/models/baseline.py`
- [Smoothed loss](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/losses/label_smoothing_torch.py): `lulc-project/src/losses/label_smoothing_torch.py`

### Training and prediction

- [Run configuration](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/config.py): `lulc-project/src/config.py`
- [Training loop](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/training/train.py): `lulc-project/src/training/train.py`
- [Cross-validation runner](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/run_cross_validation.py): `lulc-project/scripts/run_cross_validation.py`
- [Tile prediction](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/predict.py): `lulc-project/scripts/predict.py`

### Evaluation and reliability

- [Temperature scaling](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/calibration/temperature_scaling.py): `lulc-project/src/calibration/temperature_scaling.py`
- [Classification metrics](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/evaluation/metrics.py): `lulc-project/src/evaluation/metrics.py`
- [Statistical tests](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/evaluation/statistical_tests.py): `lulc-project/src/evaluation/statistical_tests.py`
- [Variant comparison](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/compare_variants.py): `lulc-project/scripts/compare_variants.py`
- [Selective prediction](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/selective_prediction.py): `lulc-project/scripts/selective_prediction.py`
- [Learned fusion gate](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/train_fusion_gate.py): `lulc-project/scripts/train_fusion_gate.py`
- [Grad-CAM](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/src/explainability/gradcam.py): `lulc-project/src/explainability/gradcam.py`
- [Grad-CAM analysis](https://github.com/sanjayimmanuel24/lulc-efficientnet-ensemble/blob/master/lulc-project/scripts/make_gradcam_figures.py): `lulc-project/scripts/make_gradcam_figures.py`

### Other components

- **Researcher**
- **EuroSAT-MS tiles**

## Connections

- Researcher → Cross-validation runner: runs folds (evidence: `README.md`)
- Researcher → Tile prediction: requests predictions (evidence: `README.md`)
- EuroSAT-MS tiles → Tile dataset: reads tiles (evidence: `lulc-project/src/data/dataset.py`)
- Tile dataset → Spectral indices: computes indices (evidence: `lulc-project/src/data/dataset.py`)
- Cross-validation runner → Run configuration: loads settings (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Cross-validation runner → Stratified splits: builds folds (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Cross-validation runner → Tile dataset: loads tiles (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Cross-validation runner → Image transforms: transforms inputs (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Cross-validation runner → Dual-branch model: trains model (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Cross-validation runner → ResNet baseline: selects baseline (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Cross-validation runner → Training loop: runs training (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Cross-validation runner → Temperature scaling: fits calibration (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Cross-validation runner → Classification metrics: computes metrics (evidence: `lulc-project/scripts/run_cross_validation.py`)
- Training loop → Smoothed loss: builds criterion (evidence: `lulc-project/src/training/train.py`)
- Dual-branch model → ECA attention: uses attention (evidence: `lulc-project/src/models/ensemble.py`)
- Dual-branch model → Torch fusion: uses fusion (evidence: `lulc-project/src/models/ensemble.py`)
- Tile prediction → Tile dataset: loads tiles (evidence: `lulc-project/scripts/predict.py`)
- Tile prediction → Image transforms: transforms inputs (evidence: `lulc-project/scripts/predict.py`)
- Tile prediction → Dual-branch model: predicts classes (evidence: `lulc-project/scripts/predict.py`)
- Tile prediction → Fusion reference: reports fusion (evidence: `lulc-project/scripts/predict.py`)
- Variant comparison → Statistical tests: tests variants (evidence: `lulc-project/scripts/compare_variants.py`)
- Selective prediction → Statistical tests: tests risk (evidence: `lulc-project/scripts/selective_prediction.py`)
- Learned fusion gate → Stratified splits: loads folds (evidence: `lulc-project/scripts/train_fusion_gate.py`)
- Learned fusion gate → Tile dataset: loads tiles (evidence: `lulc-project/scripts/train_fusion_gate.py`)
- Learned fusion gate → Dual-branch model: caches logits (evidence: `lulc-project/scripts/train_fusion_gate.py`)
- Learned fusion gate → Fusion reference: compares fusion (evidence: `lulc-project/scripts/train_fusion_gate.py`)
- Grad-CAM analysis → Grad-CAM: generates maps (evidence: `lulc-project/scripts/make_gradcam_figures.py`)
- Grad-CAM analysis → Dual-branch model: explains model (evidence: `lulc-project/scripts/make_gradcam_figures.py`)
