🛰️ Augmented Convolutional Neural Networks for Remote Sensing Change Detection

📂 Repository Index

File	Description	Model	Challenge
AML_Initial_Model_Selection.ipynb	Notebook for initial baselining and model selection	Unaugmented: EF, Siam-Conc. & Siam-Diff	Original Dataset
AML_Unaugmented.ipynb	Notebook for training unaugmented networks for few-shot learning	Unaugmented: EF	Few-shot learning
AML_Data_Augmentation.ipynb	Notebook for training augmented network with edge maps and MRA	Augmented: EF (Edge & MRA)	Original Dataset & Few-shot learning
AML_Pretrain.ipynb	Notebook for training augmented network with pretrained imagenet encoder	Augmneted: EF (Pretrained Encoder)	Original Dataset & Few-shot learning
AML_Evaluation.ipynb	Notebook for evaluating results for all phases	Unaugmented: EF, Siam-Conc. & Siam-Diff & Augmented models: EF (Edge, MRA & Pretrained)	Original Dataset, Few-shot learning, Noise Robustification & Dataset Shift
Project_Report.pdf	Report containing details about the datasets, implementation, experiments and results	---	---
README.md	Description file for the repository	---	---

✅ Instructions to run the notebooks:

1. For Phase I:

Run the cells inside the AML_Initial_Model_Selection.ipynb notebook sequentially to train and test the three Unaugmented networks (EF, Siam-Conc. & Siam-Diff.).

Use the default values in the Dataset Downloader & Dataset Characteristics cells.

2. For Phase II:

Run the cells inside the AML_Data_Augmentation.ipynb notebook sequentially to train and test Edge-Augmented & MRA-Augmented networks.

Run the cells inside the AML_Pretrain.ipynb notebook sequentially to train and test Feature-Augmented networks.

Use the default values in the Dataset Downloader & Dataset Characteristics cells. Use the default values in the Subset Choice cell.

3. For Phase III:

• Dataset Shift & Noise Robustness:

Run the first cell inside the AML_Evaluation.ipynb notebook with both the dataset choices (“LEVIRCD_Plus” & “WHU”) to download the primary and alternate dataset.

Run the second cell inside the AML_Evaluation.ipynb notebook to import requisite libraries.

Scroll to the Phase III Dataset Shift & Noise Robustness results section of the notebook, and run the cells sequentially to reproduce the results.

• Few-shot Learning:

Run the cells inside the AML_Unaugmented.ipynb notebook sequentially to train and test few-shot learning for Unaugmented EF network.

Run the cells inside the AML_Data_Augmentation.ipynb notebook sequentially to train and test few-shot learning for Edge-Augmented & MRA-Augmented EF network.

Run the cells inside the AML_Pretrain.ipynb notebook sequentially to train and test few-shot learning for Feature-Augmented EF network.

Use the default values in the Dataset Downloader & Dataset Characteristics cells.

Select respective subset percentage values (25 & 50) in the Subset choice cell for the respective experiment.

🌟 Qualitative Results

Qualitative Result I

Qualitative Result II