LeafDet: Ablation Study Results for Tomato Leaf Disease Detection Model

Vaskor Mostafa's 2026 study presents results from an ablation study of the LeafDet object detection model. The model, based on YOLOv8, was designed to detect eight distinct tomato leaf diseases using a balanced dataset of 7,836 images named PlantTom. The study reports a 91.6% [email protected] performance and includes validation via Eigen-CAM visualization.

Use Cases

• Benchmarking object detection model components based on the described ablation study results
• Training or fine-tuning disease detection models for smart agriculture based on the PlantTom dataset methodology
• Analyzing the impact of attention modules like ECA and Shuffle Attention on model accuracy and speed
• Evaluating loss functions for object detection tasks based on the reported PIoUv2 findings

Strengths

• Dataset is explicitly balanced to address class imbalance issues common in public plant disease datasets
• Model achieves a reported 91.6% [email protected], a 2.2% improvement over the baseline YOLOv8n
• Study includes interpretability validation using Eigen-CAM to visualize the decision-making process

Limitations

• Row count is unknown, which may limit suitability assessment for specific training needs
• Column-level documentation is absent; field semantics must be inferred after download
• The 9.5 KB file size suggests the dataset contains summary results, not the underlying image data

Provenance

Source

Vaskor Mostafa via figshare

Collection Method

Combined images from various public sources to create the PlantTom dataset for model training and evaluation.

Time Range

null

Freshness

Last updated 2026-05-22 18:00:33; freshness should be verified

Geography

null