Hyperparameter Tuning Results for PlantaNet Vision Models

Hyperparameter optimization results for a plant disease classification framework evaluated on a curated multi-crop dataset of approximately 45,000 images. The dataset, created by Md. Sifat Haque Zidan and last updated in April 2026, comprises 51 disease and healthy classes. Results include validation and test accuracy for eight tiny vision transformer architectures and two custom convolutional neural networks.

Use Cases

• Comparing model performance based on validation and test accuracy metrics for plant disease classification.
• Analyzing computational efficiency trade-offs based on model size and parameter counts mentioned in the description.
• Benchmarking hyperparameter optimization strategies for vision models on agricultural image data.
• Studying model interpretability based on the qualitative Grad-CAM and Grad-CAM++ analyses referenced.

Strengths

• Includes performance metrics for 10 different model architectures, such as PlantaNet achieving 99.66% test accuracy.
• Models are designed for computational efficiency, with PlantaNetLite having 1.28M parameters and a compact size of 9.85 MB.
• Dataset is based on a curated multi-crop image collection of approximately 45,000 original images across 51 classes.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The underlying image dataset's specific geographic and temporal coverage is not detailed.

Provenance

Source

figshare

Collection Method

Results from training and hyperparameter optimization of vision models on a curated multi-crop image dataset aggregated from multiple publicly available repositories.

Time Range

The results were last updated in April 2026; the temporal coverage of the underlying image data is unspecified.

Freshness

Last updated 2026-04 24 17:24:15; freshness should be verified.

Geography

The geographic coverage of the underlying plant image dataset is not specified.