Edit model card

Vision Transformer fine-tuned on Matthijs/snacks dataset

Vision Transformer (ViT) model pre-trained on ImageNet-21k and fine-tuned on Matthijs/snacks for 5 epochs using various data augmentation transformations from torchvision.

The model achieves a 94.97% and 94.43% accuracy on the validation and test set, respectively.

Data augmentation pipeline

The code block below shows the various transformations applied during pre-processing to augment the original dataset. The augmented images where generated on-the-fly with the set_transform method.

from transformers import ViTFeatureExtractor
from torchvision.transforms import (
    Compose,
    Normalize,
    Resize,
    RandomResizedCrop,
    RandomHorizontalFlip,
    RandomAdjustSharpness,
    ToTensor
)

checkpoint = 'google/vit-base-patch16-224-in21k'
feature_extractor = ViTFeatureExtractor.from_pretrained(checkpoint)

# transformations on the training set
train_aug_transforms = Compose([
    RandomResizedCrop(size=feature_extractor.size),
    RandomHorizontalFlip(p=0.5),
    RandomAdjustSharpness(sharpness_factor=5, p=0.5),
    ToTensor(),
    Normalize(mean=feature_extractor.image_mean, std=feature_extractor.image_std),
])

# transformations on the validation/test set
valid_aug_transforms = Compose([
    Resize(size=(feature_extractor.size, feature_extractor.size)),
    ToTensor(),
    Normalize(mean=feature_extractor.image_mean, std=feature_extractor.image_std),
])
Downloads last month
11
Hosted inference API
Drag image file here or click to browse from your device
This model can be loaded on the Inference API on-demand.

Dataset used to train matteopilotto/vit-base-patch16-224-in21k-snacks

Evaluation results