Context

Handwritten image data is easy to find in languages such as English and Japanese, but not for many Indian languages including Hindi. While trying to create an MNIST like personal project, I stumbled upon a Hindi Handwritten characters dataset by Shailesh Acharya and Prashnna Kumar Gyawali, which is uploaded to the UCI Machine Learning Repository.

This dataset however, only has the digits from 0 to 9, and all other characters have been removed.

Content

Data Type: GrayScale Image Image Format: PNG Resolution: 32 by 32 pixels Actual character is centered within 28 by 28 pixel, padding of 2 pixel is added on all four sides of actual character.

There are ~1700 images per class in the Train set, and around ~300 images per class in the Test set.

Acknowledgements

The Dataset is ©️ Original Authors.

Original Authors: - Shailesh Acharya - Prashnna Kumar Gyawali

Citation: S. Acharya, A.K. Pant and P.K. Gyawali “**Deep Learning Based Large Scale Handwritten Devanagari Character Recognition**”, In Proceedings of the 9th International Conference on Software, Knowledge, Information Management and Applications (SKIMA), pp. 121-126, 2015.

The full Dataset is available here: https://archive.ics.uci.edu/ml/datasets/Devanagari+Handwritten+Character+Dataset

The following dataset contains the MNIST dataset in stroke/point form. The data in this repository was based on the data obtained from the following project: https://github.com/edwin-de-jong/mnist-digits-stroke-sequence-data

Pytorch The MNIST database of handwritten digits has a training set of 60,000 examples, and a test set of 10,000 examples. It is a subset of a larger set available from NIST. The digits have been size-normalized and centered in a fixed-size image THE ORIGINAL DATA SET CAN BE FOUND IN http://yann.lecun.com/exdb/mnist/ This projects uses 2 hidden Layers with 128 and 64 units. SGD optimizer was used to improve the Weights and bias

API Tool | Project

Abstract
The dataset contains multiple sequences of aluminum foil balloon recorded by a 77GHz FMCW radar in inverse synthetic aperture (ISAR) setting.

Measurement setting
We recorded a dataset in a well-defined setting, which can be used for training ML algorithms.
For that purpose, we used a dedicated 77-GHz frequency modulated continuous wave (FMCW) radar with 2GHz bandwidth and 4 antennas forming a uniform linear array with half wavelength spacing between them.
We recorded 50 snapshots per second, where each snapshot contains one channel impulse response (CIR) per antenna with 1024 taps. The radar beam is focused by a meta-material lens, leading to four received beams covering an angle of 10° in azimuth direction [1].
Our targets are a set of foil balloons in shapes of digits from 0 to 9. Each balloon is approximately 15cm in height and its width and depth are varied from 8cm to 10cm and 3cm to 5cm respectively depending on the digit.
All measurement data are collected in an inverse synthetic aperture radar (ISAR) setting in a closed room environment.
The position and orientation of the radar are fixed through all the measurements. The digit shaped targets are placed at an initial position at the center of the radar beam at 3m distance and facing towards the radar.
In each measurement, the target is continuously rotated around its center with respect to the x-, y-, and z-axis, where the x-axis is initially pointing towards the radar and the z-axis is pointing towards the room ceiling.
The maximum rotation angle for all axes is in the range from −45° to +45° with respect to its initial orientation.
While the target is rotated, its distance towards the radar is also changed along the x-axis, in the range from −0.5m to +0.5m, relative to its initial position, but kept at the same position in the yz-plane.

Usage
The sequences are represented in NumPy arrays, stored in Pickle files that are compressed in a single Zip archive.
The corresponding meta information are stored in a Pandas Dataframe in the `dataset_meta.pkl` Pickle file.
The files can be used to filter the sequences for certain properties like label or recording environment.

For convenience, we provide an API to download and work with the dataset. The API is available at the following link: API tool

Author affiliations

References