• There are five classes.
    NameLabel =  {construction: 0, crowd: 1, pothole: 2, person-bike: 3, person-pet: 4}
  • Images are 96×96 pixels, color.
  • 50K images, each class has 10K images.
  • Images are acquired from taking pictures from the sidewalk and from the google image search.


The sidewalk obstacle data v0.1 is release under MIT license and is available to download from this link

How to load in python

The data file is available in python pkl format.

import pickle
def load_file(filename):
    '''Load the sidewalk obstacle image data'''
    with open(filename, 'rb') as f:
        datadict= pickle.load(f)
        (X_train,Y_train),(X_valid,Y_valid),(X_test,Y_test) = datadict
    	return (X_train, Y_train),(X_valid,Y_valid),(X_test,Y_test)

# load training, validation and test set 
train,valid,test = load_file('sidewalk_rgb_all.pkl')

# separate the features and label 
X_train,y_train = train
X_valid,y_valid = valid
X_test, y_test  = test


  • Please cite the following reference in papers using this dataset:
  title={Optimization and evaluation of deep architectures for ambient awareness on a sidewalk},
  author={Ahmed, Farruk and Yeasin, Mohammed},
  booktitle={Neural Networks (IJCNN), 2017 International Joint Conference on},
  • Please use http://cvpia.memphis.edu/sidewalk-obstacle-image-data/¬†as the URL when necessary


Send questions or comments to Faruk Ahmed: mf****@memphis.edu