metadata

license: apache-2.0
tags:
  - image-classification
  - computer-vision
  - vegetables
  - pytorch
  - food
datasets:
  - Custom
metrics:
  - accuracy
  - confusion_matrix
model-index:
  - name: VeggieNet
    results:
      - task:
          type: image-classification
          name: Image Classification
        dataset:
          name: Custom Vegetable Dataset
          type: image
        metrics:
          - type: accuracy
            value: 91.63%
          - type: confusion_matrix
            value: included

🥕 VeggieNet: Vegetable Image Classifier

VeggieNet is a deep learning model trained in PyTorch for classifying vegetable images into categories like tomato, carrot, potato, etc. It uses a fully connected neural network with regularization (BatchNorm and Dropout) to prevent overfitting and improve generalization.

🧠 Model Architecture

The network takes 128x128 RGB images and passes them through the following layers:

nn.Sequential(
    nn.Flatten(),
    nn.Linear(3 * 128 * 128, 512),
    nn.BatchNorm1d(512),
    nn.ReLU(),
    nn.Dropout(0.3),
    nn.Linear(512, 256),
    nn.BatchNorm1d(256),
    nn.ReLU(),
    nn.Dropout(0.3),
    nn.Linear(256, 128),
    nn.BatchNorm1d(128),
    nn.ReLU(),
    nn.Dropout(0.3),
    nn.Linear(128, num_classes)
)

Loss Function: CrossEntropyLoss
Optimizer: Adam
Input Size: 3x128x128
Output: num_classes (one per vegetable category)

📂 Dataset

This model is trained on a custom dataset from kaggle of vegetable images organized into:

vegetables_dataset/
├── train/
├── val/
└── test/

Each subfolder represents a vegetable class (e.g., carrot/, tomato/, etc.). To download Click Here

📈 Training & Evaluation

Trained for 10 epochs
Batch size: 16
Includes validation + test evaluation
Final accuracy on test set: ~91.63%
Confusion matrix is included in the evaluation

✅ Intended Use

Educational projects
Computer vision experiments
Simple food classification tasks

🚫 Limitations

Not robust to background noise or very similar vegetables
May underperform on unseen real-world data if distribution differs

💡 Future Improvements

Replace FC layers with a CNN for better spatial feature learning
Use transfer learning (e.g., ResNet18)
Increase dataset diversity and quantity

📜 License

This model is available under the Apache-2.0 License.

✍️ Author

Created by: Arun Arunisto
GitHub: arun-arunisto