Oracle-50K

Oracle Character Recognition Dataset - Oracle-50K

	Num. of Instances	Num.of Classes
Xiaoxuetang	13255	1096
Koukotsu	18671	1850
Chinese Etymology	27155	1120
Oracle-50K	59081	2668

Google Drive: https://drive.google.com/file/d/1PyXHz84viKr6N7GsWzrbPJubCkBBeLXv/view?usp=sharing

Unlabeled Source Data

The unlabeld source data for Unsupervised Few-Shot Oracle Character Recognition, including oracle and other ancient Chinese characters.
We use the online approximation algorithm from [1] to convert pixel images to sequence data.
Due to 3 failure cases during online approximation, the number of samples in pixel format and sequence format are 276,031 and 276,028, respectively.

Google Drive: https://drive.google.com/file/d/1cn-LCIs61P5I26yMHboPdSTj0MOkgcUM/view?usp=sharing

Oracle-FS

The Few-Shot Oracle dataset for Unsupervised Few-Shot Oracle Character Recognition.
Oracle-FS is constructed from Oracle-50K by randomly choose 200 categories.
k-shot: k training examples and 20 test examples per class.

Google Drive: https://drive.google.com/file/d/1EbK0zuNpUFYmfZVAq4nDmVWsy7REO4hx/view?usp=sharing\

 # A simple case for .npz file loading and sample visualization

 import numpy as np
 import random
 import matplotlib.pyplot as plt

 npz_path = 'xxx.npz'
 data = np.load(npz_path, encoding='latin1', allow_pickle=True)
 train_data = data['train']
 test_data = data['test']

 sample = np.random.choice(test_data, 1)[0]

 def strokes_to_lines(strokes):
   """Convert stroke-3 format to polyline format."""
   x = 0
   y = 0
   lines = []
   line = []

   for i in range(len(strokes)):
       if strokes[i, 2] == 1:
           x += float(strokes[i, 0])
           y += float(strokes[i, 1])
           line.append([x, y])
           lines.append(line)
           line = []
       else:
           x += float(strokes[i, 0])
           y += float(strokes[i, 1])
           line.append([x, y])
   return lines

 def show_one_sample(strokes, linewidth=100):
   lines = strokes_to_lines(strokes)
   for idx in range(0, len(lines)):
       x = [x[0] for x in lines[idx]]
       y = [y[1] for y in lines[idx]]
       plt.plot(x, y, 'k-', linewidth=linewidth)

   ax = plt.gca()
   plt.xticks([])
   plt.yticks([])
   ax.xaxis.set_ticks_position('top')
   ax.invert_yaxis()
  
  # show
  plot(sample)