Example for loading the splits used in "Exploring Compositional High Order Pattern Potentials for Structured Output Learning" from Li et al.

weizman_dataset.py

# TODO: Get weizmann_32_32_trainval.mat from https://www.cs.toronto.edu/~yujiali/papers/chopps.zip
# TODO: Get horse.mat                   from https://www.cs.toronto.edu/~yujiali/papers/cvpr13_data.zip
# TODO: Update the paths below
split_path='YOUR_PATH_TO/weizmann_32_32_trainval.mat'
data_path='YOUR_PATH_TO/horse.mat'

# Imports
import matplotlib.pyplot as plt
import numpy as np
import scipy.io
import json

split_info=scipy.io.loadmat(split_path)
image_data = scipy.io.loadmat(data_path)

def get_data_for_split(data, split_info, split_name):
  """Get the images and masks for a given split."""
  splits = {"train": [0, 160], "val": [160, 200], "test": [200, 328]}
  split = splits[split_name]
  # Substract 1, as the idx uses 1-based indexing.
  indices= split_info['idx'][0][split[0]:split[1]] - 1
  images = [i[0] for i in data['imgs'][indices]]
  masks = [m.reshape(32,32).T for m in data['segs'][indices]]
  
  return images, masks

# Load the data for a given split ('train','val' or 'test')
images, masks = get_data_for_split(image_data, split_info, 'test')

# Show the first image
plt.subplot(1,2,1)
plt.imshow(masks[-1])
plt.grid('off')
plt.subplot(1,2,2)
plt.imshow(images[-1])
plt.grid('off')