README.md

metadata

library_name: transformers
license: apache-2.0
base_model: google/vit-base-patch16-224
tags:
  - image-classification
  - food
  - fruits
  - junkfood
  - generated_from_trainer
metrics:
  - accuracy
model-index:
  - name: vit-food-classification-gutkia01
    results: []

vit-food-classification-gutkia01

This model is a fine-tuned version of google/vit-base-patch16-224 on the food-classification dataset. It achieves the following results on the evaluation set:

• Loss: 0.0008
• Model Preparation Time: 0.0034
• Accuracy: 0.9998

🍎 Fruits vs. Junkfood Classifier – Vision Transformer (gutkia01)

This model is a fine-tuned version of google/vit-base-patch16-224, trained on a custom binary dataset to distinguish between healthy fruits and unhealthy fast food.

🧠 Model Description

• Architecture: Vision Transformer (ViT)
• Base model: google/vit-base-patch16-224
• Task: Binary image classification: Fruit vs. Junkfood
• Framework: Hugging Face Transformers Trainer
• Input format: RGB images, 224×224, loaded via imagefolder

✅ Intended Use & Limitations

Appropriate Use Cases

• Food classification in nutrition, health, or educational applications
• Interactive demos comparing healthy vs. unhealthy food
• Computer vision use cases with simple binary class structures

Limitations

• Only supports binary classification (no subclass differentiation)
• Cannot recognize new or abstract dishes (e.g. salad, sushi)
• Cannot evaluate ingredients, calories, or portion sizes

📊 Training and Evaluation Data

The model was trained on a binary dataset composed of:

• Fruits360 Dataset: 137,000+ structured fruit images in a controlled studio setup (Kaggle link)
• Fast Food Classification Dataset v2: 20,000 fast food images, 10 categories (e.g., burger, pizza, fries) (Kaggle link)

Dataset Composition

The dataset is a combination of:

• Fruits360 Dataset (Kaggle)
• Fast Food Classification Dataset v2 (Kaggle)

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 6

Training results

Epoch	Training Loss	Validation Loss	Accuracy
1	0.0000	0.0215	0.9975
2	0.0000	0.00004	1.0000
3	0.0000	0.00008	1.0000
4	0.0000	0.00011	1.0000
5	0.0000	0.00011	1.0000
6	0.0000	0.00008	1.0000

Final training loss: 0.00047
Evaluation accuracy: 0.9998

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 2.14.4
• Tokenizers 0.21.1