def __init__(self, opts, params):
self.opts = opts
self.dataset_params = params
self.data_dir = params.data_dir
self.dtype = params.dtype
self.n_views = params.views[-1]
self.n_pts = params.points[-1]
d = self.n_pts * self.n_views
e = params.descriptor_dim
p = params.points[-1]
f = opts.final_embedding_dim
self.features = {
'InitEmbeddings':
tf_helpers.TensorFeature(
key='InitEmbeddings',
shape=[d, e],
dtype=self.dtype,
description='Initial embeddings for optimization'),
'AdjMat':
tf_helpers.TensorFeature(key='AdjMat',
shape=[d, d],
dtype=self.dtype,
description='Adjacency matrix for graph'),
'Degrees':
tf_helpers.TensorFeature(key='Degrees',
shape=[d, d],
dtype=self.dtype,
description='Degree matrix for graph'),
'Laplacian':
tf_helpers.TensorFeature(
key='Laplacian',
shape=[d, d],
dtype=self.dtype,
description='Alternate Laplacian matrix for graph'),
'TrueEmbedding':
tf_helpers.TensorFeature(
key='TrueEmbedding',
shape=[d, p],
dtype=self.dtype,
description='True values for the low dimensional embedding'),
'NumViews':
tf_helpers.Int64Feature(
key='NumViews',
description='Number of views used in this example'),
'NumPoints':
tf_helpers.Int64Feature(
key='NumPoints',
description='Number of points used in this example'),
}
def __init__(self, opts, params):
super(GraphSimDataset, self).__init__(opts, params)
d = self.n_pts * self.n_views
self.features['Mask'] = \
tf_helpers.TensorFeature(
key='Mask',
shape=[d, d],
dtype=self.dtype,
description='Mask for valid values of matrix')
self.features['MaskOffset'] = \
tf_helpers.TensorFeature(
key='MaskOffset',
shape=[d, d],
dtype=self.dtype,
description='Mask offset for loss')
def __init__(self, opts, params):
super().__init__(opts, params)
self.features.update({
'rots':
tf_helpers.TensorFeature(key='rots',
shape=[self.tuple_size, 3, 3],
dtype=self.dtype,
description='Rotations of the cameras'),
'trans':
tf_helpers.TensorFeature(
key='trans',
shape=[self.tuple_size, 3],
dtype=self.dtype,
description='Translations of the cameras'),
})
def __init__(self, opts, params):
self.opts = opts
self.dataset_params = params
self.data_dir = params.data_dir
self.dtype = params.dtype
p = params.points[-1]
v = params.views[-1]
self.n_views = v
self.n_pts = p
d = p * v
e = params.descriptor_dim
# GraphsTuple(nodes=nodes,
# edges=edges,
# globals=globals,
# receivers=receivers,
# senders=senders,
# n_node=n_node,
# n_edge=n_edge)
self.features = {
'n_node':
tf_helpers.Int64Feature(
key='n_node',
dtype='int32',
description='Number of nodes we are using'),
'nodes':
tf_helpers.TensorFeature(
key='nodes',
shape=[d, e],
dtype=self.dtype,
description='Initial embeddings for optimization'),
'n_edge':
tf_helpers.Int64Feature(
key='n_edge',
dtype='int32',
description='Number of edges in this graph'),
'globals':
tf_helpers.VarLenFloatFeature(key='globals',
shape=[None],
description='Edge features'),
'edges':
tf_helpers.VarLenFloatFeature(key='edges',
shape=[None, 1],
description='Edge features'),
'receivers':
tf_helpers.VarLenIntListFeature(
key='receivers',
dtype='int32',
description='Recieving nodes for edges'),
'senders':
tf_helpers.VarLenIntListFeature(
key='senders',
dtype='int32',
description='Sending nodes for edges'),
}
def __init__(self, opts, params):
super(GeomKNNRome16KDataset,
self).__init__(opts, params, tuple_size=params.views[-1])
d = self.n_pts * self.n_views
e = params.descriptor_dim
self.features.update({
'InitEmbeddings':
tf_helpers.TensorFeature(
key='InitEmbeddings',
shape=[d, e + 2 + 1 + 1],
dtype=self.dtype,
description='Initial embeddings for optimization'),
'Rotations':
tf_helpers.TensorFeature(key='Rotations',
shape=[self.tuple_size, 3, 3],
dtype=self.dtype,
description='Mask offset for loss'),
'Translations':
tf_helpers.TensorFeature(key='Translations',
shape=[self.tuple_size, 3],
dtype=self.dtype,
description='Mask offset for loss'),
})
def __init__(self, opts, params):
"""
Inputs:
- opts (options) - object with all relevant options stored
- params (DatasetParams) - object with all dataset parameters stored
Outputs: GraphSimDataset
"""
self.opts = opts
self.dataset_params = params
self.data_dir = params.data_dir
self.dtype = params.dtype
self.n_views = np.random.randint(params.views[0], params.views[1] + 1)
self.n_pts = np.random.randint(params.points[0], params.points[1] + 1)
d = self.n_pts * self.n_views
e = params.descriptor_dim
p = params.points[-1]
f = opts.final_embedding_dim
self.features = {
'InitEmbeddings':
tf_helpers.TensorFeature(
key='InitEmbeddings',
shape=[d, e],
dtype=self.dtype,
description='Initial embeddings for optimization'),
'AdjMat':
tf_helpers.TensorFeature(key='AdjMat',
shape=[d, d],
dtype=self.dtype,
description='Adjacency matrix for graph'),
'Degrees':
tf_helpers.TensorFeature(key='Degrees',
shape=[d, d],
dtype=self.dtype,
description='Degree matrix for graph'),
'Laplacian':
tf_helpers.TensorFeature(
key='Laplacian',
shape=[d, d],
dtype=self.dtype,
description='Alternate Laplacian matrix for graph'),
'Mask':
tf_helpers.TensorFeature(
key='Mask',
shape=[d, d],
dtype=self.dtype,
description='Mask for valid values of matrix'),
'MaskOffset':
tf_helpers.TensorFeature(key='MaskOffset',
shape=[d, d],
dtype=self.dtype,
description='Mask offset for loss'),
'TrueEmbedding':
tf_helpers.TensorFeature(
key='TrueEmbedding',
shape=[d, p],
dtype=self.dtype,
description='True values for the low dimensional embedding'),
'NumViews':
tf_helpers.Int64Feature(
key='NumViews',
description='Number of views used in this example'),
'NumPoints':
tf_helpers.Int64Feature(
key='NumPoints',
description='Number of points used in this example'),
}