metadata

license: apache-2.0
datasets:
  - ylecun/mnist
language:
  - en
base_model:
  - google/siglip2-base-patch16-224
pipeline_tag: image-classification
library_name: transformers
tags:
  - Digits
  - Mnist
  - SigLIP2
  - 0-t0-9
  - Number-Classification

Mnist-Digits-SigLIP2

Mnist-Digits-SigLIP2 is an image classification model fine-tuned from google/siglip2-base-patch16-224 to classify handwritten digits (0-9) using the SiglipForImageClassification architecture. It is trained on the MNIST dataset for accurate digit recognition.

Classification Report:
              precision    recall  f1-score   support

           0     0.9988    0.9959    0.9974      5923
           1     0.9987    0.9918    0.9952      6742
           2     0.9918    0.9943    0.9930      5958
           3     0.9975    0.9938    0.9957      6131
           4     0.9892    0.9882    0.9887      5842
           5     0.9859    0.9937    0.9898      5421
           6     0.9936    0.9939    0.9937      5918
           7     0.9856    0.9943    0.9899      6265
           8     0.9932    0.9921    0.9926      5851
           9     0.9926    0.9897    0.9912      5949

    accuracy                         0.9928     60000
   macro avg     0.9927    0.9928    0.9927     60000
weighted avg     0.9928    0.9928    0.9928     60000

Classes:

Class 0: "0"
Class 1: "1"
Class 2: "2"
Class 3: "3"
Class 4: "4"
Class 5: "5"
Class 6: "6"
Class 7: "7"
Class 8: "8"
Class 9: "9"

Run with Transformers🤗

!pip install -q transformers torch pillow gradio

import gradio as gr
from transformers import AutoImageProcessor, SiglipForImageClassification
from transformers.image_utils import load_image
from PIL import Image
import torch

# Load model and processor
model_name = "prithivMLmods/Mnist-Digits-SigLIP2"
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

def classify_digit(image):
    """Predicts the digit in the given handwritten digit image."""
    image = Image.fromarray(image).convert("RGB")
    inputs = processor(images=image, return_tensors="pt")
    
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist()
    
    labels = {
        "0": "0", "1": "1", "2": "2", "3": "3", "4": "4",
        "5": "5", "6": "6", "7": "7", "8": "8", "9": "9"
    }
    predictions = {labels[str(i)]: round(probs[i], 3) for i in range(len(probs))}
    
    return predictions

# Create Gradio interface
iface = gr.Interface(
    fn=classify_digit,
    inputs=gr.Image(type="numpy"),
    outputs=gr.Label(label="Prediction Scores"),
    title="MNIST Digit Classification 🔢",
    description="Upload a handwritten digit image (0-9) to recognize it using MNIST-Digits-SigLIP2."
)

# Launch the app
if __name__ == "__main__":
    iface.launch()

Sample Inference

Intended Use:

The Mnist-Digits-SigLIP2 model is designed for handwritten digit recognition. Potential applications include:

Optical Character Recognition (OCR): Digit recognition for various documents.
Banking & Finance: Automated check processing.
Education & Learning: AI-powered handwriting assessment.
Embedded Systems: Handwriting input in smart devices.