Training a pedestrian detection for a self-driving car

The dataset for training a pedestrian object detection can be found at http://pascal.inrialpes.fr/data/human/. The steps to detecting pedestrians can be found at https://github.com/diegocavalca/machine-learning/blob/master/supervisioned/object.detection_tensorflow/simple.detection.ipynb. The dataset for training a Sign Detector can be downloaded from http://www.vision.ee.ethz.ch/~timofter/traffic_signs/ and http://btsd.ethz.ch/shareddata/. In the case of a self-driving car, there would be four classes in an image for labeling: pedestrian, car, motorcycle, and background. The background class has to be detected when none of the classes is present. An assumption in training a deep learning classification model is that at least one of the objects will be present in the image. By adding the background class, we are overcoming the problem. The neural network can also produce a bounding box of the object from the label.