cifar10_vit16_lora

Model Details

Model Description

An adapter for the google/vit-base-patch16-224 ViT trained on CIFAR10 classification task

Loading guide

from transformers import AutoModelForImageClassification

labels2title = ['plane', 'car', 'bird', 'cat',
    'deer', 'dog', 'frog', 'horse', 'ship', 'truck']
model = AutoModelForImageClassification.from_pretrained(
    'google/vit-base-patch16-224-in21k',
    num_labels=len(labels2title),
    id2label={i: c for i, c in enumerate(labels2title)},
    label2id={c: i for i, c in enumerate(labels2title)}
)
model.load_adapter("yturkunov/cifar10_vit16_lora")

Learning curves

Recommendations to input

The model expects an image that has went through the following preprocessing stages:

• Scaling range: <img src="https://latex.codecogs.com/gif.latex?[0, 255]\rightarrow[0, 1]" />
• Normalization parameters: <img src="https://latex.codecogs.com/gif.latex?\mu=(.5,.5,.5),\sigma=(.5,.5,.5)" />
• Dimensions: 224x224
• Number of channels: 3

Inference on 3x4 random sample