-
Notifications
You must be signed in to change notification settings - Fork 1
/
create_complete_graphs.py
298 lines (246 loc) · 10.3 KB
/
create_complete_graphs.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
import matplotlib.pyplot as plt
import numpy as np
import torch
import pandas as pd
from torch import nn
from torch import optim
import torch.nn.functional as F
from torchvision import transforms
import torchvision.models as models
import matplotlib as mpl
mpl.use('Agg')
from torch.autograd import Variable
from PIL import Image
import os, glob
import csv
from torch_geometric.data import InMemoryDataset
from torch_geometric.data import Data, DataLoader
import argparse
import os
from scipy.spatial import distance
# creating my own resnet model
class myResnetModel(nn.Module):
def __init__(self):
super(myResnetModel, self).__init__()
self.layer1 = child_list[0]
self.layer2 = child_list[1]
self.layer3 = child_list[2]
self.layer4 = child_list[3]
self.layer5 = child_list[4]
self.layer6 = child_list[5]
self.layer7 = child_list[6]
self.layer8 = child_list[7]
# adding my own conv layer in pretrained resnet-18
self.layer9 = nn.Conv2d(512, 16, 5)
# adding fc layers for classification
self.fc1 = nn.Linear(256, 64)
self.fc2 = nn.Linear(64, 2)
#self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
def forward(self, x):
out1 = self.layer1(x)
out1 = self.layer2(out1)
out1 = self.layer3(out1)
out1 = self.layer4(out1)
out1 = self.layer5(out1)
out1 = self.layer6(out1)
out1 = self.layer7(out1)
out1 = self.layer8(out1)
out1 = self.layer9(out1)
out1 = out1.view(-1, self.flat_features(out1))
out1 = F.relu(self.fc1(out1))
out1 = F.dropout(out1, training=self.training, p=0.7)
out1 = F.log_softmax(self.fc2(out1), dim=1)
#out1 = self.avgpool(out1)
return out1
def flat_features(self, x):
size = x.size()[1:] # all dimensions except batch dimensions
num_features = 1
for s in size:
num_features *= s
return num_features
# read images from command line
#pic_one = str(input("Input first image name\n"))
#pic_two = str(input("Input second image name\n"))
# automate
def parse_args():
parser = argparse.ArgumentParser(description="Create Graph data structure")
parser.add_argument('-l', '--layer', type=str, required=True, help='give layer of resnet-18')
parser.add_argument('-p0', '--path0', type=str, required=True, help='give path of class 0')
parser.add_argument('-p1', '--path1', type=str, required=True, help='give path of class 1')
parser.add_argument('-s', '--saveGraph', type=str, required=True, help='path to store graph data')
parser.add_argument('wt', 'model', type=str, required=True, help='path to the model' )
parser.add_argument('-n', '--name', type=str, required=True, help='graph name')
return parser
#rcv inps from cmd
if __name__ == "__main__":
parser = parse_args()
args = parser.parse_args()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# load model
#'/home/sachin/Desktop/Guided_research/scrpit_4_r18/task1_myresnet/myresnet_weights.pth'
model=torch.load(args.model)
#model = models.resnet18(pretrained=True)
print(model)
# select desired layer which basically creates reference to the layer we want to extract from
#layer = model._modules.get('layer9')
layer = model._modules.get(args.layer)
print(layer)
#set to eval mode so as to ensure that any dropout layers are not active during forward pass
model.eval()
# img transform
scaler = transforms.Resize((256, 256))
normalize = transforms.Normalize(mean=[0.29874189944720375, 0.5893857441207402, 0.9193030837391296],
std = [0.14290491468798067, 0.14531491548727832, 0.0992852656427492])
to_tensor = transforms.ToTensor()
def get_feature_embeddings(image_name):
# load img with Pillow library
img = Image.open(image_name)
# create a pytorch var with transformed image
# unsqueeze reshape img from (3, 224, 224) to (1, 3, 224, 224) since pytorch expects 4-D input
t_img = Variable(normalize(to_tensor(scaler(img))).unsqueeze(0))
t_img = t_img.to(device)
# create vector of zeros holds feature vector
# avg pool layer
#my_embedding = torch.zeros(256, 16, 16)
my_embedding = torch.zeros(16, 4, 4)
# fn that copy o/p of layer
""" m=module, i=grad input, o=grad ouytput """
def copy_data(m, i, o):
my_embedding.copy_(o.data.squeeze())
# attach above fn to our selected layer
h = layer.register_forward_hook(copy_data)
# run model on transformed img
model(t_img)
# detach copy fn from layer
h.remove()
#return feature embedding vector
return my_embedding
# testing with 1 img
#print(get_feature_embeddings('test1.jpg').shape)
#e = get_feature_embeddings('test2.jpg')
#e = e[4, :, : ].flatten()
#print('e shape', e.shape)
# holds input graph data for GNN
data_list = []
# store graph data in list
count = 0
def create_pytorch_geom_dataset(node_features, edge_index, edge_features, label):
global count
count = count +1
data = Data(x=node_features, edge_index=edge_index, edge_attr=edge_features, y=label)
print('Converting img data into graph: ', data, 'img count:', count)
data_list.append(data)
# create edge index of graph basically adjacency matrix
def create_edge_index(ids_list, id_count):
source_nodes = []
target_nodes = []
# create source node indexes
for i in range(id_count):
for j in range(id_count-1):
source_nodes.append(ids_list[i].item())
# create target node indexes
for i in range(id_count):
if i == 0:
target_nodes.append(ids_list[1:])
elif i == id_count:
target_nodes.append(ids_list[:-1])
else:
lhs = ids_list[:i]
rhs = ids_list[i+1:]
join_list = lhs + rhs
target_nodes.append(join_list)
# flatten list
target_nodes = [item for sublist in target_nodes for item in sublist]
edge_index = torch.tensor([source_nodes,
target_nodes], dtype=torch.long)
return edge_index, source_nodes, target_nodes
# minx =-1
# maxx = 1
# def calculate_cosine_sim(fv1, fv2):
# fv1 = fv1.detach().cpu().numpy()
# fv2 = fv2.detach().cpu().numpy()
# dot_product = np.dot(fv1,fv2)
# norm_a = np.linalg.norm(fv1)
# norm_b = np.linalg.norm(fv2)
# return ((dot_product/(norm_a * norm_b)-minx)/(maxx-minx))
def correlation_distance(feature_v1, feature_v2, node_matrix):
feature_v1 = feature_v1.detach().cpu().numpy()
feature_v2 = feature_v2.detach().cpu().numpy()
# calculate pairwise distances
p_dist = distance.pdist(node_matrix, metric='correlation')
# convert to a square symmetric distance matrix
sq_dist = distance.squareform(p_dist)
sigma = np.mean(sq_dist)
# calculate correlation distance between each pair of the feature vector
distv = distance.cdist(feature_v1, feature_v2, metric='correlation')
# compute simlarity measure using correlation distance and sigma
sim = np.exp(-distv**2 / (2 * sigma**2))
return sim
# create edge feature using cosine similarity between two feature embeddings
def create_edge_features(edge_index, node_matrix, source_nodes, target_nodes):
num_edge_features = edge_index.shape[1]
edge_features = torch.zeros([num_edge_features, 1], dtype=torch.float)
for idx in range(num_edge_features):
#print(edge_index[:, idx])
feature_vector_pair = edge_index[:, idx]
v1_idx = feature_vector_pair[0]
v2_idx = feature_vector_pair[1]
feature_v1 = node_matrix[v1_idx]
feature_v1 = feature_v1.to(device)
feature_v2 = node_matrix[v2_idx]
feature_v2 = feature_v2.to(device)
#feature_v1 = torch.tensor(feature_v1, dtype=torch.float)
#feature_v2 = torch.tensor(feature_v2, dtype=torch.float)
corel_dist = correlation_distance(feature_v1.unsqueeze(0), feature_v2.unsqueeze(0), node_matrix)
corel_dist = torch.FloatTensor(corel_dist)
edge_features[idx] = corel_dist
return edge_features
def form_graph_node_features(feature_embeddings):
# layer 9
node_feature_matrix = torch.zeros([16, 16], dtype=torch.float)
for idx in range(feature_embeddings.shape[0]):
flatten_features = feature_embeddings[idx, :, :].flatten()
node_feature_matrix[idx] = flatten_features
return node_feature_matrix
def create_graph_data_structre(inp_img, label):
feature_embeddings = get_feature_embeddings(inp_img)
node_feature_matrix = form_graph_node_features(feature_embeddings)
print('node feature matrix shape', node_feature_matrix.shape)
node_index_list = torch.arange(0, node_feature_matrix.shape[0])
node_index_list = list(node_index_list)
edge_index, source_nodes, target_nodes = create_edge_index(node_index_list, node_feature_matrix.shape[0])
edge_features = create_edge_features(edge_index, node_feature_matrix, source_nodes, target_nodes)
print('edge index shape', edge_index.shape)
# flatten edge features
edge_features = edge_features.flatten()
print('edge features shape', edge_features.shape)
create_pytorch_geom_dataset(node_feature_matrix, edge_index, edge_features, label)
# test input
#create_graph_data_structre('test1.jpg', 0)
# read data dir with respective class path
#path_class_0 = '/b_test/sharma/Guided_Research/Origa650/origa/class_0/'
path_class_0 = args.path0
for file in os.listdir(path_class_0):
full_path = path_class_0+file
create_graph_data_structre(full_path, 0)
#path_class_1 = '/b_test/sharma/Guided_Research/Origa650/origa/class_1/'
path_class_1 = args.path1
for file in os.listdir(path_class_1):
full_path = path_class_1+file
create_graph_data_structre(full_path, 1)
class MyBinaryDataset(InMemoryDataset):
def __init__(self, root, transform=None, pre_transform=None):
super(MyBinaryDataset, self).__init__(root, transform, pre_transform)
self.data, self.slices = torch.load(self.processed_paths[0])
@property
def raw_file_names(self):
return []
@property
def processed_file_names(self):
return[args.name] #return ['/b_test/sharma/Guided_Research/Optic_Discs']
def download(self):
pass
def process(self):
data, slices = self.collate(data_list)
torch.save((data, slices), self.processed_paths[0])
dataset = MyBinaryDataset(root=args.saveGraph)