metadata
base_model: WinKawaks/vit-tiny-patch16-224
datasets:
- 0-ma/geometric-shapes
license: apache-2.0
metrics:
- accuracy
pipeline_tag: image-classification
Model Card for VIT Geometric Shapes Dataset Tiny
Training Dataset
Base Model
Accuracy
- Accuracy on dataset 0-ma/geometric-shapes [test] : 0.9138095238095238
Loading and using the model
import numpy as np
from PIL import Image
from transformers import AutoImageProcessor, AutoModelForImageClassification
import requests
labels = [
"None",
"Circle",
"Triangle",
"Square",
"Pentagon",
"Hexagon"
]
images = [Image.open(requests.get("https://raw.githubusercontent.com/0-ma/geometric-shape-detector/main/input/exemple_circle.jpg", stream=True).raw),
Image.open(requests.get("https://raw.githubusercontent.com/0-ma/geometric-shape-detector/main/input/exemple_pentagone.jpg", stream=True).raw)]
feature_extractor = AutoImageProcessor.from_pretrained('0-ma/vit-geometric-shapes-tiny')
model = AutoModelForImageClassification.from_pretrained('0-ma/vit-geometric-shapes-tiny')
inputs = feature_extractor(images=images, return_tensors="pt")
logits = model(**inputs)['logits'].cpu().detach().numpy()
predictions = np.argmax(logits, axis=1)
predicted_labels = [labels[prediction] for prediction in predictions]
print(predicted_labels)
Model generation
The model has been created using the 'train_shape_detector.py.py' of the project from the project https://github.com/0-ma/geometric-shape-detector. No external code sources were used.