• This project recognizes Chinese characters in a image file and convert them into Arabic numerals
  Check demo.ipynb for the whole process
• Now the seperation is done with simple matrix multiplication labeling. To handle rotated input, use Spectral Clustering for number seperation. Define the similarity matrix with gaussian kernal and run the K-mean on the eigenspace.

Pytorch
OpenCV
torchvision

• Original photo

• After process

Chinese MNIST

• demo.ipynb a example to show how the program works.
• src/
  • model.py Defination of class IntRec, the model trained to convert chinese character to Arabic numerals.
  • utils.py
    • Implementation of DataLoader, which load image files from dataset into tensor of shape
      [batch, channel, height, width] along with groundtruth label
    • HelperFunc a class of helper function to extract single characters from each image
  • train.py traning loops for the model, aumomatically use cuda if availiable. Tracking training and validation accuracy in train_acc.pickle and valid_acc.pickle
• data/ containing training data images and a csv file chinese_mnist.csv, which contains paths to images and labels.
• pictures/ the program will read image files from this folder and produce prediction

• Chen Li - Initial work -