CLOVES-4603: Benchmarking Classical Texture Features and Fine-Tuned Deep Models for Clove Quality Grading
Authors: Innocent Nyalala, Patrick Vincent Ndowo
Affiliation: SAAIL Lab, IIT Madras Zanzibar Campus, Tanzania
Venue: 7th Workshop on Vision for Agriculture (V4A), CVPR 2026, Denver, USA
We introduce CLOVES-4603, the first public benchmark dataset for automated clove quality grading, comprising 4,603 standardized 224×224 RGB images spanning four commercial grades collected at the Zanzibar State Trading Corporation (ZSTC). We benchmark classical texture pipelines (HOG+SVM, GLCM+GLRLM+LBP fusion) against fine-tuned deep learning models including ResNet-50, MobileNetV3, and a frozen DINOv2 vision transformer. ResNet-50 achieves 99.67% accuracy on the test set, while our texture-fusion SVM reaches 92.94%, providing a strong and deployment-friendly classical baseline. This benchmark is designed to inform real-world deployment trade-offs between accuracy, latency, and computational cost in agricultural settings where infrastructure is limited.
| Model | Accuracy | Notes |
|---|---|---|
| ResNet-50 (fine-tuned) | 99.67% | Best overall |
| MobileNetV3-Large (fine-tuned) | 99.35% | Lightweight option |
| DINOv2 ViT-B/14 (frozen) | 94.90% | Zero-shot baseline |
| GLCM+GLRLM+LBP SVM | 92.94% | Classical baseline |
| HOG+SVM | 32.68% | Weak baseline |
CLOVES-4603 contains 4,603 images across four official ZSTC commercial grades:
| Grade | Description |
|---|---|
| Grade 1 | Attractive golden/saffron color, max 3% mpeta |
| Grade 2 | Faded/slightly blackish, max 7% mpeta |
| Grade 3 | More faded color, max 20% mpeta |
| Grade 4 | Primarily mpeta cloves, >20% mpeta |
Dataset download link coming soon via Zenodo.
CLOVES-4603/
├── data/ # Dataset placeholder (see Zenodo link above)
├── src/
│ ├── classical/ # Texture feature pipelines
│ ├── deep/ # Fine-tuning scripts
│ └── evaluation/ # Metrics and visualization
├── notebooks/ # Experiment notebooks
├── requirements.txt
└── README.md
git clone https://github.com/saaillab/CLOVES-4603
cd CLOVES-4603
pip install -r requirements.txt@inproceedings{nyalala2026cloves,
title={CLOVES-4603: Benchmarking Classical Texture Features and Fine-Tuned Deep Models for Clove Quality Grading},
author={Nyalala, Innocent and Ndowo, Patrick Vincent},
booktitle={7th Workshop on Vision for Agriculture (V4A), CVPR},
year={2026}
}SAAIL Lab (Sustainable AI for Agriculture and Intelligent Livelihoods) is based at IIT Madras Zanzibar Campus, Tanzania. We build responsible, locally grounded AI solutions for East Africa and the Global South.