Convolutional Neural Network to classify images of birds into Cardinal and Indigo. Only TensorFlow operations are used to read image files, and build CNN.
TensorFlow 1.0, Numpy
tf.layers, tf.image, tf.gfile, tf.losses
2 Convolutional layers with ReLU activation, 2 Max Pooling Layers, 1 Fully Connected Layer with Softmax Activation
Training data: 72 images, Validation data: 24 images, Testing data: 4 images
0: Cardinal; 1: Indigo; 2: None of these
Source: CUB-200 available at http://www.vision.caltech.edu/visipedia/CUB-200.html