spaceml-org/fireclr-wildfires

How spaceml-org/fireclr-wildfires works

FireCLR is a research workflow for self-supervised wildfire change detection from four-band satellite imagery. Users prepare image scenes as tiles, train a SimCLR-based encoder, then compare pre- and post-fire representations to produce cosine-distance maps and evaluate them against annotated burn-severity labels. The graph includes the sampled model and training utilities plus the README-documented preprocessing and downstream stages. Notebook internals were not sampled, so links to their detailed runtime components are left unasserted.

Main entry points: simclr.py SimCLR trainer · Preparing_Training_Data_S2.ipynb Training tile preparation · Preparing_Val_Data.ipynb Validation tile preparation · FireCLR_Downstream_CD_S2.ipynb Downstream inference notebooks

Components 17

Data preparation

Training tile preparation notebooks/preprocessing/Preparing_Training_Data_S2.ipynb
Validation tile preparation notebooks/preprocessing/Preparing_Val_Data.ipynb
Training tile loader utils/load_data.py

Contrastive training

View generation utils/ContrastiveLearning.py
Image augmentations utils/augmentation.py
SimCLR trainer models/simclr.py
FireCLR encoder models/cnn_simclr.py
CNN backbone models/simple_cnn.py
Checkpoint and metrics utils/save_models.py
Sensor training notebooks Mesa_FireCLR_Training_S2.ipynb

Change detection

Downstream inference notebooks FireCLR_Downstream_CD_S2.ipynb
Cosine distance maps notebooks/downstream_task/S2_Downstream_validation.ipynb

Validation

Burn severity evaluation notebooks/downstream_task/validation.py
Validation notebooks notebooks/downstream_task/S2_Downstream_validation.ipynb

Other components

Research user
Satellite scenes
Burn severity labels

Connections 17

Research user → Training tile preparation: prepares tiles README.md
Satellite scenes → Training tile preparation: provides scenes README.md
Research user → Validation tile preparation: prepares scenes README.md
Satellite scenes → Validation tile preparation: provides scenes README.md
Training tile preparation → Sensor training notebooks: prepares training data README.md
Sensor training notebooks → SimCLR trainer: runs training
Training tile loader → View generation: transforms samples
View generation → Image augmentations: applies transforms
SimCLR trainer → FireCLR encoder: encodes batches models/simclr.py
FireCLR encoder → CNN backbone: delegates encoding models/cnn_simclr.py
SimCLR trainer → Checkpoint and metrics: saves checkpoints models/simclr.py
Research user → Downstream inference notebooks: runs inference README.md
Validation tile preparation → Downstream inference notebooks: prepares validation tiles README.md
Downstream inference notebooks → Cosine distance maps: produces distance maps README.md
Cosine distance maps → Validation notebooks: evaluates predictions README.md
Burn severity labels → Validation notebooks: supplies labels README.md
Validation notebooks → Burn severity evaluation: computes metrics

More Jupyter Notebook diagrams: orpheus-ai/zeus · rahulsajnani/geodiffuser · rikluost/ofdm-plutosdr-pytorch · yusuke-imoto-lab/recode · chlotmpo/pathcore_anomaly_detection

Drawn by GitDiagram on October 1, 2026 from the repository's file tree, README and source files. Read as Markdown ·