Held-out test set
COMPUTER VISION
A closer look.
A lighter model.
Exploring the balance between detection quality and computational complexity for nine paddy disease classes.
AIT2309246803 test images2,466 annotated objects
Held-out test set
3,905,069 parameters
Reported model summary
DETECTION STUDIO
One image. Three perspectives.
Compare your baselines and proposed model on the same image.
Upload a paddy image in the live demo to view each model’s labeled predictions. The Render service may take about a minute to wake after inactivity.
THE EXPERIMENTS
Performance, in perspective.
Explore the main comparison and ablation study.
Loading recorded results…
Detection performance
Performance–complexity trade-off
| Model | Precision (%) | Recall (%) | mAP@0.5 (%) | mAP@0.5:0.95 (%) | Parameters (M) | GFLOPs |
|---|
CLASS-LEVEL DETECTION PERFORMANCE
Every class has a story.
Loading nine classes…
PROPOSED METHOD
DenseLiteX + SPPF + C2PSA
A custom backbone studied within the YOLOv8n detection framework. The ablation study compares the DenseLiteX backbone and the contributions of SPPF and C2PSA.
Paddy imageFeature extractionMulti-scale detectionDisease labels + boxes
Charts show recorded test-set results. Uploaded-image predictions are available in the live demo. GFLOPs come from model summaries; timings are not reported here.