train_person_classification.py

# import IN THIS ORDER - otherwise cv2 gets loaded after tensorflow,
# and tensorflow loads an incompatible internal version of libpng
# https://github.com/tensorflow/tensorflow/issues/1924
import cv2
import numpy as np
import tensorflow as tf

from Datasets.Dataset import batcher

from Datasets.tud import load_tud
from Datasets.inria import load_inria
from Datasets.zurich import load_zurich

import Model

class PersonModel(Model.BooleanModel):
    def train(self, person_iter):
        batch_size = 50
        for batch_no, batch in enumerate(batcher(person_iter, batch_size=100)):
            train_accuracy = self.accuracy.eval(feed_dict={
                self.x:batch[0], self.y_: batch[1], self.keep_prob: 1.0})
            if batch_no % 5 == 0:
                print("Step %i, training accuracy %g"%(batch_no, train_accuracy))
                # r = y_conv.eval(feed_dict={self.x: batch[0], keep_prob: 1.0})
                # print('Guess: ',  np.round(r.flatten()))
                # print('Actual:', np.round(batch[1].flatten()))
            self.train_step.run(feed_dict={self.x: batch[0], self.y_: batch[1], self.keep_prob: 0.5})

if __name__ == '__main__':
    combined_dataset = load_tud('/mnt/data/Datasets/pedestrians/tud/tud-pedestrians') + \
          load_tud('/mnt/data/Datasets/pedestrians/tud/tud-campus-sequence') + \
          load_tud('/mnt/data/Datasets/pedestrians/tud/tud-crossing-sequence') + \
          load_tud('/mnt/data/Datasets/pedestrians/tud/TUD-Brussels') + \
          load_tud('/mnt/data/Datasets/pedestrians/tud/train-210') + \
          load_tud('/mnt/data/Datasets/pedestrians/tud/train-400') + \
          load_tud('/mnt/data/Datasets/pedestrians/tud/TUD-MotionPairs/positive') + \
          load_tud('/mnt/data/Datasets/pedestrians/tud/TUD-MotionPairs/negative') + \
          load_inria('/mnt/data/Datasets/pedestrians/INRIA/INRIAPerson') + \
          load_zurich('/mnt/data/Datasets/pedestrians/zurich')

    combined_dataset.train.generate_negative_examples()
    combined_dataset.test.generate_negative_examples()
    combined_dataset.shuffle()
    combined_dataset.balance()

    train_pos = combined_dataset.train.num_positive_examples
    train_neg = combined_dataset.train.num_negative_examples
    print(len(combined_dataset.train), 'training examples ({},{}).'.format(train_pos, train_neg))
    print(len(combined_dataset.test), 'testing examples ({},{}).'.format(combined_dataset.test.num_positive_examples, combined_dataset.test.num_negative_examples))

    nn_im_w = 64
    nn_im_h = 160

    with tf.Session() as sess:
        model = PersonModel(sess)
        model.build_graph(nn_im_w, nn_im_h)

        print("Training...")
        model.train(combined_dataset.train.iter_people())

        model.save('saved_model/')