thermography.classification.models package

The models (computation graphs) which can be used to train a classifier are coded here.

Submodules

thermography.classification.models.base_net module

class BaseNet(x: tensorflow.python.framework.ops.Tensor, image_shape: numpy.ndarray, num_classes: int, name: str = 'ThermoNet')

Bases: abc.ABC

Base interface for nets used by the thermography package. This class offers a convenience interface to train a classifier and to access its data.

channels

Returns the number of channels accepted by the model.

create() → None

Method which each subclass must implement which creates the computational graph.

Note

This method must use the input placeholder self.x and terminate into the logits tensor self.logits.

flat_shape

Returns the shape of the image which the model takes as input to the computational graph.

Example:

img_batch.shape = [batch_shape, w, h, c]
flat = model.flat_shape # [w, h]

image_shape

Returns the image shape accepted by the model. This shape corresponds to the shape of each element carried by the input placeholder self.x.

logits

Returns a tf.Tensor representing the logits of the classified input images.

name

Returns the name associated to the model.

num_classes

Returns the number of classes which the model will consider for classification.

static update_shape(current_shape: numpy.ndarray, scale: int) → numpy.ndarray

Updates the current shape of the data in the computational graph when a pooling operation is applied to it.

Example:

current_shape = self.flat_shape # [w, h]
conv = conv_relu(x=self.x, kernel_shape, bias_shape)
pool = max_pool_2x2(name="max_pool", x=conv)
current_shape = self.update_shape(current_shape, 2) # [w/2, h/2]

x

Returns a reference to the input placeholder.

thermography.classification.models.thermo_net module

class ThermoNet(x: tensorflow.python.framework.ops.Tensor, image_shape: numpy.ndarray, num_classes: int, keep_prob: float, *args, **kwargs)

Bases: thermography.classification.models.base_net.BaseNet

create()

thermography.classification.models.thermo_net_3x3 module

class ThermoNet3x3(x: tensorflow.python.framework.ops.Tensor, image_shape: numpy.ndarray, num_classes: int, keep_prob: float, *args, **kwargs)

Bases: thermography.classification.models.base_net.BaseNet

Class representing the computational graph structure used for classification of solar panel modules.

Initializes the computational graph according to the structure defined in self.create.

Parameters:

• x – Tensorflow placeholder which will be fed with a batch of images.
• image_shape – Shape of each input image fed to self.x.
• num_classes – Number of classes which the model can predict.
• keep_prob – Tensorflow placeholder representing the keep probability for the dropout layers.

create()

Creates the ThermoNet3x3 computation graph consisting of three convolutional layers followed by three fully connected layers.

Thermonet3x3 Structure