Markov Random Field for Image Segmentation in Crop Monitoring
Overview
What this challenge is about.
You receive 60 Sentinel-2 image tiles (10-meter resolution) over 12 vineyards, each tile with per-pixel disease labels from agronomist field walks. Take the consultancy's existing random-forest posteriors as given (you do NOT need to retrain). Define a 4-connected MRF over the pixel grid with an Ising-style smoothness prior and a per-pixel data term from the random forest. Run inference with graph cuts or iterated conditional modes. Validate on 10 held-out tiles using both pixel-level intersection-over-union (IoU) and field-level F1 (a vineyard counts as positive if more than 5 percent of pixels are flagged). Success is a field-level F1 lift of at least 8 points over the raw random forest.
The Brief
What you'll do, and what you'll demonstrate.
Add an MRF smoothness layer over an existing per-pixel classifier to deliver field-level disease maps the agronomists trust.
Earning criteria — what you'll demonstrate
- Formulate spatial smoothing as inference on a pairwise Markov Random Field
- Apply graph-cut or ICM (Iterated Conditional Modes) inference on a real image grid
- Tune a smoothness prior strength via cross-validation on held-out tiles
- Communicate the lift in a way a non-ML agronomist client can verify
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Applied AI Scientist
Layering a classical probabilistic model on top of an existing ML pipeline to fix a real client trust issue is the bread and butter of applied AI work at small consultancies.
This challenge sharpens
- markov-random-fields
- graph-cuts
- image-segmentation
Computer Vision Engineer
Spatial inference on image grids is a transferable CV-engineer skill that shows up in medical imaging, satellite analytics, and document understanding.
This challenge sharpens
- image-segmentation
- spatial-modeling
- graph-cuts
Machine Learning Engineer
Owning a post-processing pipeline with documented validation and tuning hand-off is the kind of MLE follow-through that earns trust on a team.
This challenge sharpens
- python
- model-evaluation
- spatial-modeling