USPS Dataset

USPS Dataset

A database for handwritten text recognition research.

Download Raw Dataset

USPS Dataset

USPS Testing Dataset

Dataset Statistics

  1. # of classes: 10
  2. # of data: 7291 / 2007 (testing)
  3. # of features: 256

Samples

USPS Sample

Dataset Usage

TensorFlow:

https://github.com/darshanbagul/USPS_Digit_Classification

USPS_Digit_Classification

Requirements
  tensorflow==1.0.1
  matplotlib==1.5.1
  cv2==2.4.13
  numpy==1.13.0
  cPickle==1.71
Implementation
1. Logistic Regression

2. Single layered Neural Networks

3. Convolutional Neural Networks
Results
  1. Logistic Regression

    Accuracy on USPS data - 36.30 %

  2. Single Layer Neural Network

    Accuracy on USPS data - 49.23 %

  3. Convolutional Neural Networks

    Accuracy on USPS data - 63.0 %

Usage:
git clone https://github.com/darshanbagul/USPS_Digit_Classification.git
cd USPS_Digit_Classification
python main.py