MIIA Pothole Image Classification Challenge
Identify which images of roads in South Africa contain potholes
248 data scientists enrolled, 45 on the leaderboard
ConstructionTransportationSafetyComputer VisionImageUnstructured
South Africa
2 September 2019

The data contains images of streets in South Africa that either have potholes (positive response) or do not (negative response). The data have been split into a test and train set. The training set contains 4,026 images of test set contains 1,650 images.

Your task is to provide the probability that an image contains a pothole. For each unique image ID you should estimate the likelihood that the image contains a pothole, with an estimated probability value between 0 and 1.

The files you have for download here are:

  • sample_submission.csv: is an example of what your submission file should look like, including a list of unique image IDs. Your submission file should have all of the Image IDs in this file along with estimates of the corresponding probabilities of observing a pothole. The order of the Image IDs does not matter, but the names need to be correct.
  • train_ids_labels.csv: list of file names of images labeled as pothole (1) or no pothole (0) that make up the training set, i.e. these are the images on which you should train your model.
  • test_ids_only.csv: list of file names of images that are not labeled. This is the test set on which you should apply your model.
  • all_data.zip: all image files (both test and train)

Get started on this challenge with this blog by Johnowhitaker.