Skip to content

model-benchmarking.md

Latest commit

 

History

History
79 lines (53 loc) · 5.31 KB

model-benchmarking.md

File metadata and controls

79 lines (53 loc) · 5.31 KB
Raw

Model Performance & Benchmarking

This page centralizes the model metrics and reproducibility notes that are currently spread across the README, training notebooks, and exported result artifacts.

Benchmark Summary

Crop Task Model Primary metrics Evidence artifacts
Cotton Growth stage detection YOLOv8 mAP50 60.06%, mAP95 34.8%, recall 53.8%, precision 62.7%, inference 3.3 ms results/cotton_crop_growth_stage_prediction/results.csv, results/cotton_crop_growth_stage_prediction/results.png, confusion matrices
Cotton Disease classification ResNet50 accuracy 99.83%, precision 99.83%, recall 99.83%, F1 99.83%, ROC AUC 99.98% results/cotton_crop_disease_classification/test_metrics.txt, accuracy/loss curves, confusion matrix
Tomato Disease detection YOLOv8 mAP50 95.4%, mAP95 86.2%, recall 88.5%, precision 92.3%, inference 1.3 ms results/tomato_crop_disease_classification/results.csv, results/tomato_crop_disease_classification/results.png, PR/F1 curves
Tomato Growth stage classification ResNet50 accuracy 100%, precision 100%, recall 100%, F1 100% results/tomato_crop_growth_stage_classification/classification_report.txt, accuracy/loss curves, confusion matrix
Potato Disease classification CNN notebook pipeline see classification report artifact results/potato_crop_disease_classification/classification_report.txt, accuracy/loss curves, confusion matrix

Confusion Matrix Artifacts

Use the exported matrices to inspect class-level error patterns instead of relying on aggregate accuracy alone.

Cotton Disease Classification

Cotton disease confusion matrix

Cotton Growth Stage Detection

Cotton growth stage confusion matrix

Tomato Growth Stage Classification

Tomato growth stage confusion matrix

Potato Disease Classification

Potato disease confusion matrix

Training Curves

Training curves help identify overfitting, underfitting, and unstable validation behavior.

Task Accuracy / metrics Loss / training result
Cotton disease classification Cotton disease accuracy curve Cotton disease loss curve
Cotton growth stage detection Cotton growth metrics Cotton growth training results
Tomato disease detection Tomato disease training results Tomato disease F1 curve
Tomato growth stage classification Tomato growth accuracy curve Tomato growth loss curve
Potato disease classification Potato disease accuracy curve Potato disease loss curve

Sample Prediction Artifacts

The repository currently includes exported tomato disease detection samples. Keep sample outputs in this section tied to checked-in artifacts so reviewers can trace each screenshot back to the run that produced it.

Task Sample outputs
Tomato disease detection Tomato disease sample prediction 1 Tomato disease sample prediction 2

Reproducing Results

The checked-in notebooks under scripts/ are the source of truth for reproducing the exported artifacts:

Task Notebook
Cotton disease classification scripts/cotton_crop_disease_prediction.ipynb
Cotton growth stage detection scripts/cotton_crop_growth_stage_prediction.ipynb
Tomato disease detection scripts/tomato-crop-disease-classification.ipynb
Tomato growth stage classification scripts/tomato-growth-stages-classification.ipynb
Potato disease classification scripts/potato_crop_disease_classification.ipynb

Recommended reproducibility checklist:

  1. Download the dataset linked in the README for the target crop/task.
  2. Keep the same train/validation/test split used by the notebook.
  3. Run the corresponding notebook from scripts/.
  4. Export the generated metrics, confusion matrix, and curves into the matching results/<task>/ directory.
  5. Update the benchmark summary only when the artifact files and metric values come from the same run.

Notes for Future Model Updates

  • Do not update headline metrics without committing or linking the matching evaluation artifact.
  • Prefer per-class reports and confusion matrices over accuracy alone, especially for imbalanced disease classes.
  • Include inference-time measurements when changing model size, image resolution, or preprocessing.