Course Roadmap Mapping
This week’s work in the broader GFM plan.
| Week | Stage | Focus | You will build (geogfm) | Library tools | Outcome |
|---|---|---|---|---|---|
| 10 | Stage 3: Apply & Deploy | Presentations & Synthesis | Project deliverables (no new geogfm code required) |
— | Present MVP builds, analysis, transition plan |
Weekly goals
- Synthesize architecture, training, and evaluation learnings
- Present MVP results and insights
- Outline next steps with Prithvi/scale-up
Session Outline
- Concepts → Components mapping
- Model architecture →
models/gfm_vit.py - Pretraining objective →
models/gfm_mae.py+modules/losses/mae_loss.py - Training loop →
training/{optimizer.py, loop.py} - Evaluation →
evaluation/{visualization.py, metrics.py} - Applied tasks →
tasks/{classification.py, segmentation.py} - Inference →
inference/{sliding_window.py, tiling.py}
- Model architecture →
- Deliverables checklist
- Slides with pipeline diagram and key code references
- Short demo: load a small batch, run MAE forward, show recon
- One analysis figure (recon grid or PSNR curve)
- What you would swap to scale (timm blocks, TorchGeo datasets, FlashAttention, HF hub)