zutaars1/vit-computer-vision-classification-model

vit-base-mini-food-3

This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the Mini Food-3 custom dataset. It achieves the following results on the evaluation set:

• Loss: 0.5656
• Accuracy: 0.9067

Model description

This is a Vision Transformer (ViT) model fine-tuned for food image classification. The model was trained to classify three food categories: pizza, sushi, and ice cream.

Intended uses & limitations

This model is intended for classifying images of three food types:

• pizza
• sushi
• ice_cream

It performs best on images similar to the training dataset (Food-101 subset). The model may not generalize well to other food categories or different image conditions.

Training and evaluation data

The model was trained on the Mini Food-3 dataset, derived from the Food-101 dataset:

Split	Images per class	Total images
Train	100	300
Validation	25	75
Test	25	75
Total	150	450

Preprocessing

• Images converted to RGB
• Resizing and normalization using ViT image processor
• Automatic label encoding from folder structure

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.00002
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: adamw_torch with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3
• weight_decay: 0.01

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.9543	1.0	38	0.8170	0.8667
0.6416	2.0	76	0.6260	0.9067
0.4879	3.0	114	0.5656	0.9067

Framework versions

• Transformers 4.50.0
• PyTorch 2.0+
• Datasets 2.0+
• Tokenizers 0.13+

Model Comparison

This fine-tuned ViT model was compared against:

• CLIP (Zero-Shot): openai/clip-vit-large-patch14 - No training required
• OpenAI Vision Model: LLM-based image classification

The fine-tuned ViT model outperforms both baseline approaches on this specific dataset due to task-specific training.