easy_vision.python.core.optical_flow.tvnet

easy_vision.python.core.optical_flow.tvnet.spatial_transformer

easy_vision.python.core.optical_flow.tvnet.spatial_transformer.batch_transformer(U, thetas, out_size, name='BatchSpatialTransformer')

Batch Spatial Transformer Layer

Parameters:

• U (float) – tensor of inputs [num_batch,height,width,num_channels]
• thetas (float) – a set of transformations for each input [num_batch,num_transforms,6]
• out_size (int) – the size of the output [out_height,out_width]
• Returns (float) – Tensor of size [num_batch*num_transforms,out_height,out_width,num_channels]

easy_vision.python.core.optical_flow.tvnet.spatial_transformer.transformer(U, theta, out_size, name='SpatialTransformer', **kwargs)

Spatial Transformer Layer

Implements a spatial transformer layer as described in [1]. Based on [2] and edited by David Dao for Tensorflow.

Parameters:

• U (float) – The output of a convolutional net should have the shape [num_batch, height, width, num_channels].
• theta (float) – The output of the localisation network should be [num_batch, 6].
• out_size (tuple of two ints) – The size of the output of the network (height, width)

References

[1]	Spatial Transformer Networks Max Jaderberg, Karen Simonyan, Andrew Zisserman, Koray Kavukcuoglu Submitted on 5 Jun 2015

[2]	https://github.com/skaae/transformer_network/blob/master/transformerlayer.py

Notes

To initialize the network to the identity transform init theta to :

identity = np.array([[1., 0., 0.],

[0., 1., 0.]])

identity = identity.flatten() theta = tf.Variable(initial_value=identity)

easy_vision.python.core.optical_flow.tvnet.tvnet

class easy_vision.python.core.optical_flow.tvnet.tvnet.TVNet

Bases: object

GRAD_IS_ZERO = 1e-12

__init__()

x.__init__(…) initializes x; see help(type(x)) for signature

centered_gradient(x, name)

divergence(x, y, name)

dual_tvl1_optic_flow(x1, x2, u1, u2, tau=0.25, lbda=0.15, theta=0.3, warps=5, max_iterations=5)

forward_gradient(x, name)

gaussian_smooth(x)

get_loss(x1, x2, tau=0.25, lbda=0.15, theta=0.3, warps=5, zfactor=0.5, max_scales=5, max_iterations=5)

grey_scale_image(x)

normalize_images(x1, x2)

tvnet_flow(x1, x2, tau=0.25, lbda=0.15, theta=0.3, warps=5, zfactor=0.5, max_scales=5, max_iterations=5)

warp_image(x, u, v)

zoom_image(x, new_height, new_width)

zoom_size(height, width, factor)