-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathdataset.py
205 lines (161 loc) · 8.65 KB
/
dataset.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
import os
import skimage
import skimage.transform
import matplotlib.pyplot as plt
import numpy as np
from torch.utils.data import Dataset
import torch
from config import config
import torch.nn.functional as F
from new_utils import (remove_intersection_and_duplicate, sort_graph, render,
get_wrong_corners, get_wrong_edges, simplify_gt,
get_corner_bin_map, get_corner_label, get_edge_label, alpha_blend)
MAX_DATA_STORAGE = 20000 #1500
class EvaluatorDataset(Dataset):
def __init__(self, datapath, phase='train', edge_strong_constraint=False):
super(EvaluatorDataset, self).__init__()
self.datapath =datapath
self.database = []
self.new_data = []
self.ground_truth = {}
self.edge_strong_constraint = edge_strong_constraint
name = os.path.join(self.datapath, '{}_list.txt'.format(phase))
with open(name, 'r') as f:
self.namelist = f.read().splitlines()
# load original result
self.ip_datapath = os.path.join(self.datapath, 'data/ip')
self.convmpn_datapath = os.path.join(self.datapath, 'data/conv-mpn')
self.peredge_datapath = os.path.join(self.datapath, 'data/per_edge')
self.gt_datapath = os.path.join(self.datapath, 'data/gt')
for idx, name in enumerate(self.namelist):
if os.path.exists(os.path.join(self.convmpn_datapath, name+'.npy')):
conv_data = np.load(os.path.join(self.convmpn_datapath, name+'.npy'), allow_pickle=True).tolist()
conv_data['corners'], conv_data['edges'] = \
remove_intersection_and_duplicate(conv_data['corners'], conv_data['edges'], name)
conv_data['corners'], conv_data['edges'] = sort_graph(conv_data['corners'], conv_data['edges'])
corners = conv_data['corners']
edges = conv_data['edges']
gt_data = np.load(os.path.join(self.gt_datapath, name+'.npy'), allow_pickle=True).tolist()
gt_data['corners'], gt_data['edges'] = sort_graph(gt_data['corners'], gt_data['edges'])
self.ground_truth[name] = gt_data
self.add_data(name, corners, edges)
if os.path.exists(os.path.join(self.ip_datapath, name+'.npy')):
conv_data = np.load(os.path.join(self.ip_datapath, name+'.npy'), allow_pickle=True).tolist()
conv_data['corners'] = np.round(conv_data['corners']).astype(int)
conv_data['corners'], conv_data['edges'] = sort_graph(conv_data['corners'], conv_data['edges'])
corners = conv_data['corners']
edges = conv_data['edges']
self.add_data(name, corners, edges)
if os.path.exists(os.path.join(self.peredge_datapath, name+'.npy')):
conv_data = np.load(os.path.join(self.peredge_datapath, name+'.npy'), allow_pickle=True).tolist()
conv_data['corners'] = np.round(conv_data['corners']).astype(int)
conv_data['corners'], conv_data['edges'] = sort_graph(conv_data['corners'], conv_data['edges'])
corners = conv_data['corners']
edges = conv_data['edges']
self.add_data(name, corners, edges)
def __len__(self):
self.merged_data = self.database + self.new_data
return len(self.merged_data)
def __getitem__(self, idx):
data = self.merged_data[idx]
name = data['name']
corners = data['corners']
edges = data['edges']
corner_false_id = data['corner_false_id']
edge_false_id = data['edge_false_id']
img_orig = skimage.img_as_float(plt.imread(os.path.join(self.datapath, 'rgb', name+'.jpg')))
img = img_orig.transpose((2,0,1))
img = (img - np.array(config['mean'])[:, np.newaxis, np.newaxis]) / np.array(config['std'])[:, np.newaxis, np.newaxis]
mask = render(corners, edges, render_pad=-1, scale=1)
noise = torch.rand(corners.shape)*4-2 #[-2,2]
corners = corners + noise.numpy()
### corner ###
corner_gt_mask = render(corners[corner_false_id], np.array([]), render_pad=0, scale=1)[1:]
### edge ###
edge_correct_id = list(set(np.arange(edges.shape[0])) - set(edge_false_id))
edge_gt_mask = render(corners, edges[list(edge_false_id)], render_pad=0, scale=1)[0:1]
out_data = {}
gt_data = self.ground_truth[name]
gt_corners = gt_data['corners']
gt_edges = gt_data['edges']
heat_map = render(gt_corners, gt_edges, render_pad=0, corner_size=5, edge_linewidth=3)
heat_map = torch.FloatTensor(heat_map)
out_data['gt_heat_map'] = heat_map
img = torch.FloatTensor(img)
mask = torch.FloatTensor(mask)
corner_gt_mask = torch.FloatTensor(corner_gt_mask)
edge_gt_mask = torch.FloatTensor(edge_gt_mask)
out_data['img'] = img
out_data['mask'] = mask
out_data['corner_gt_mask'] = corner_gt_mask
out_data['edge_gt_mask'] = edge_gt_mask
out_data['name'] = name
return out_data
def make_data(self, name, corners, edges):
gt_data = self.ground_truth[name]
corner_false_id, map_same_degree, map_same_location = get_wrong_corners(
corners, gt_data['corners'], edges, gt_data['edges'])
gt_corners, gt_edges = simplify_gt(map_same_location, gt_data['corners'], gt_data['edges'])
corner_false_id, map_same_degree, map_same_location = get_wrong_corners(
corners, gt_corners, edges, gt_edges)
if self.edge_strong_constraint:
edge_false_id = get_wrong_edges(
corners, gt_corners, edges, gt_edges,
map_same_degree)
else:
edge_false_id = get_wrong_edges(
corners, gt_corners, edges, gt_edges,
map_same_location)
return {'name': name, 'corners': corners, 'edges': edges,
'corner_false_id': list(corner_false_id),
'edge_false_id': edge_false_id}
def add_processed_data(self, data):
self.database.append(data)
def add_data(self, name, corners, edges):
return self.add_processed_data(self.make_data(name, corners, edges))
def _add_processed_data_(self, data):
if len(self.new_data) >= MAX_DATA_STORAGE:
del self.new_data[0]
self.new_data.append(data)
def _add_data_(self, name, corners, edges):
return self._add_processed_data_(self.make_data(name, corners, edges))
class myDataset(Dataset):
def __init__(self, datapath, phase='train'):
super(myDataset, self).__init__()
self.datapath = datapath
self.phase = phase
self.database = []
name = os.path.join(self.datapath, phase+'_list.txt')
with open(name, 'r') as f:
namelist = f.read().splitlines()
# load conv-mpn result
conv_mpn_datapath = os.path.join(self.datapath, 'data/conv-mpn')
gt_datapath = os.path.join(self.datapath, 'data/gt')
self.name2id = {}
print("load conv-mpn result")
for idx, name in enumerate(namelist):
if os.path.exists(os.path.join(conv_mpn_datapath, name+'.npy')):
conv_data = np.load(os.path.join(conv_mpn_datapath, name+'.npy'), allow_pickle=True).tolist()
conv_data['corners'], conv_data['edges'] = \
remove_intersection_and_duplicate(conv_data['corners'], conv_data['edges'], name)
conv_data['corners'], conv_data['edges'] = sort_graph(conv_data['corners'], conv_data['edges'])
gt_data = np.load(os.path.join(gt_datapath, name+'.npy'), allow_pickle=True).tolist()
gt_data['corners'], gt_data['edges'] = sort_graph(gt_data['corners'], gt_data['edges'])
self.database.append({'conv_data': conv_data, 'gt_data': gt_data,
'name': name, 'corner_data': None, 'edge_data': None,
'region_data': None})
self.name2id[name] = len(self.database)-1
print("done.......")
def __len__(self):
return len(self.database)
def getDataByName(self, name):
return self.database[self.name2id[name]]
def __getitem__(self, idx):
name = self.database[idx]['name']
img = skimage.img_as_float(plt.imread(os.path.join(self.datapath, 'rgb', name+'.jpg')))
img = img.transpose((2,0,1))
img = (img - np.array(config['mean'])[:, np.newaxis, np.newaxis]) / np.array(config['std'])[:, np.newaxis, np.newaxis]
data = {
'img': img,
'name': name}
return data