CSIRO Image2Biomass Prediction.ipynb

michaelml.dev / Michael_Portfolio.ipynb

Kaggle competition for predicting pasture biomass from images using a 4-model stacking ensemble. Achieved 21% MAE improvement over the best individual model with R² of 0.9998.

[ ]

code · CSIRO Image2Biomass Prediction

project = load_project("csiro-image2biomass")
project.summary()

Output

CSIRO Image2Biomass Prediction

Kaggle competition for predicting pasture biomass from images using a 4-model stacking ensemble. Achieved 21% MAE improvement over the best individual model with R² of 0.9998.

PyTorchDINOv2ConvNeXtSigLIPStacking EnsembleCatBoostComputer VisionRegressionKaggleAgricultural AI

[ ]

code · Problem definition

project = load_project("csiro-image2biomass")
project.problem()

Output

ProblemPasture biomass traditionally requires destructive sampling; the competition asked models to estimate five biomass targets from images and metadata.

ApproachStack DINOv2, ConvNeXt, SigLIP/boosting, and DINOv2-Giant/Lasso pipelines with image augmentation, metadata features, TTA, and grouped cross-validation.

OutcomeRecorded MAE 10.03, a 21.37% improvement over the best individual model in the project evaluation, with weighted R² 0.9998.

[ ]

code · Constraints

project = load_project("csiro-image2biomass")
project.constraints()

Output

Constraints

Five related regression targets had different error profiles.
Repeated image paths made leakage-safe grouping essential.
The solution had to combine image and structured metadata features.

[ ]

code · Role and implementation

project = load_project("csiro-image2biomass")
project.implementation()

Output

Role and implementation

My role: Designed and evaluated the four-model vision ensemble, feature pipeline, grouped validation strategy, and Lasso stacking meta-model.

Combined DINOv2, ConvNeXt, SigLIP, boosting models, and Lasso in a stacking ensemble.
Used Albumentations, Mixup/CutMix, test-time augmentation, and metadata encoders.
Grouped five-fold validation by image path.

[ ]

code · Architecture

project = load_project("csiro-image2biomass")
project.show_architecture()

Output

Architecture overview

Stack DINOv2, ConvNeXt, SigLIP/boosting, and DINOv2-Giant/Lasso pipelines with image augmentation, metadata features, TTA, and grouped cross-validation.

The documented project does not claim a separate architecture artifact.

[ ]

code · Demo and media

project = load_project("csiro-image2biomass")
project.media()

Output

CSIRO Image2Biomass Prediction project media

[ ]

code · Evaluation

project = load_project("csiro-image2biomass")
project.evaluation()

Output

Evaluation

Compared all four base models against the stacking meta-model.
Tracked MAE and per-target as well as weighted R².

[ ]

code · Results

project = load_project("csiro-image2biomass")
project.results()

Output

Results

Stacking MAE: 10.03 versus 12.76 for the best individual model.
Weighted R²: 0.9998 in the documented project evaluation.

[ ]

code · Tradeoffs and failures

project = load_project("csiro-image2biomass")
project.tradeoffs()

Output

Tradeoffs

The ensemble improves aggregate error while increasing inference cost and operational complexity.
Aggregate R² can obscure weaker performance on individual biomass targets.

[ ]

code · What I would improve

project = load_project("csiro-image2biomass")
project.improvements()

Output

Improvements

Report uncertainty and additional held-out robustness checks by geography and species.
Distill the ensemble for more practical field inference.

[ ]

code · Links and repository

project = load_project("csiro-image2biomass")
project.links()

Output

GitHub

Role and implementation

My role: Designed and evaluated the four-model vision ensemble, feature pipeline, grouped validation strategy, and Lasso stacking meta-model.

Combined DINOv2, ConvNeXt, SigLIP, boosting models, and Lasso in a stacking ensemble.

Used Albumentations, Mixup/CutMix, test-time augmentation, and metadata encoders.

Grouped five-fold validation by image path.