PlantaNet: Plant Disease Classification Model Performance Summary

99.66% test accuracy was achieved by the PlantaNet model on a multi-crop dataset of approximately 45,000 images. The dataset, created by Md. Sifat Haque Zidan, aggregates 51 disease and healthy classes from public repositories. This Excel file, last updated in April 2026, contains efficiency summaries for eight tiny vision transformers and two custom convolutional neural networks.

Use Cases

• Benchmarking lightweight vision transformer architectures based on the described validation performance and computational overhead.
• Comparing custom convolutional neural network designs based on the reported parameter counts, model sizes, and accuracy metrics.
• Analyzing model interpretability based on the qualitative Grad-CAM and Grad-CAM++ analyses mentioned in the description.
• Evaluating training strategies for plant disease classification based on the described use of controlled augmentation and 100-epoch training.

Strengths

• Includes performance metrics for 10 different model architectures, including validation and test accuracy up to 99.66%.
• Reports specific model parameters (e.g., 1.28M and 2.58M) and file sizes (e.g., 9.85 MB).
• Based on a curated dataset of approximately 45,000 original images across 51 classes.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The 5.5 KB file size suggests the dataset is a summary table, not the underlying image data.

Provenance

Source

Md. Sifat Haque Zidan

Collection Method

Aggregated from multiple publicly available image repositories and curated.

Time Range

null

Freshness

Last updated 2026-04-24 17:24:14; freshness should be verified

Geography

null