CoM: all Combinations of Missing views in the Earth observation domain

Public repository of our work Missing Data as Augmentation in the Earth Observation Domain: A Multi-View Learning Approach

---

The previous image illustrates our FCoM approach in a Multi-view learning model implementing Feature-level fusion. We focus on multi-view Earth observation applications, including classification and regression tasks.

Training

We provide config file examples on how to train our model with different dynamic merge functions. The examples are on the CropHarvest data.

• To train the feature-level CoM with average fusion:

python train.py -s config/com_average.yaml

• To train the feature-level CoM with gated fusion:

python train.py -s config/com_gated.yaml

• To train the feature-level CoM with cross-attention fusion:

python train.py -s config/com_cross.yaml

• To train the feature-level CoM with memory-based fusion:

python train.py -s config/com_memory.yaml

Note

Read about the used data in data folder

Evaluation

• To evaluate the predictive performance:

python evaluate_predictions.py -s config/evaluation.yaml

• To evaluate the performance robustness:

python evaluate_rob_pred.py -s config/evaluation.yaml

---

Baselines

• To train ensemble based aggregation with sensor invariant (Reference)

python train_ensemble_baseline.py -s config/baseline_ensemble.yaml

• To train input fusion with temporal dropout (Reference)

python train_input_baseline.py -s config/baseline_inputTempD.yaml

• To train input fusion with sensor dropout (Reference)

python train_input_baseline.py -s config/baseline_inputSensD.yaml

• To train feature fusion with sensor dropout (Reference)

python train.py -s config/baseline_featureSensD.yaml

• To train feature-level concatenation fusion with CoM (Reference)

python train.py -s config/baseline_featureCoM.yaml

---

🖊️ Citation

Mena, Francisco, et al. "Missing Data as Augmentation in the Earth Observation Domain: A Multi-View Learning Approach." Neurocomputing, 2025.

@article{mena2025missing,
  title = {Missing data as augmentation in the Earth Observation domain: A multi-view learning approach},
  author = {Mena, Francisco and Arenas, Diego and Dengel, Andreas},
  journal = {Neurocomputing},
  volume = {638},
  year = {2025},
  issn = {0925-2312},
  doi = {10.1016/j.neucom.2025.130175},
  publisher={Elsevier},
}