Datasets:
Commit
•
bd00799
1
Parent(s):
5fe3c1e
updated init.py
Browse files- __init__.py +96 -40
__init__.py
CHANGED
@@ -1,43 +1,96 @@
|
|
1 |
-
import torch
|
2 |
-
import torchvision
|
3 |
-
from torchvision import transforms
|
4 |
-
from torch import Tensor
|
5 |
|
6 |
-
import pandas as
|
7 |
-
from skimage import io
|
8 |
-
import matplotlib.pyplot as
|
9 |
-
from pathlib import Path
|
|
|
10 |
|
11 |
# Ignore warnings
|
12 |
-
import warnings
|
13 |
-
|
14 |
|
15 |
-
|
16 |
-
|
17 |
-
class MORRIS(
|
18 |
_storage_csv='morris.csv'
|
19 |
_storage_jpg='jpgs'
|
20 |
-
def __init__(self,root=
|
21 |
-
self.storage =
|
22 |
self.transform = transform
|
23 |
-
self.index =
|
24 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
def __len__(self):
|
26 |
return len(self.index)
|
27 |
-
|
28 |
def __getitem__(self,idx):
|
29 |
item = self.index.iloc[idx].to_dict()
|
30 |
-
image =
|
31 |
-
|
32 |
if self.transform:
|
33 |
image = self.transform(image)
|
34 |
-
|
35 |
item['image'] = image
|
36 |
return item
|
37 |
-
|
38 |
-
def
|
39 |
-
|
40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
def _self_validate(self):
|
42 |
"""try loading each image in the dataset"""
|
43 |
allgood=True
|
@@ -48,50 +101,53 @@ class MORRIS(torch.utils.data.Dataset):
|
|
48 |
allgood=False
|
49 |
print(f"couldn't load {self.index.iloc[idx].filename}")
|
50 |
if allgood:
|
51 |
-
print(f"All
|
52 |
|
53 |
class Deframe(object):
|
54 |
"""check for uniform color boundaries on edges of input and crop them away"""
|
55 |
-
|
|
|
56 |
def __init__(self,aggressive=False,maxPixelFrame=20):
|
57 |
self.alpha = 0.1 if aggressive else 0.01
|
58 |
self.maxPixelFrame = maxPixelFrame
|
59 |
-
|
60 |
def _map2idx(self,frameMap):
|
61 |
try:
|
62 |
return frameMap.tolist().index(False)
|
63 |
except ValueError:
|
64 |
return self.maxPixelFrame
|
65 |
-
|
66 |
def _Border(self,img: Tensor):
|
67 |
""" take greyscale Tensor
|
68 |
return left,right,top,bottom border size identified """
|
|
|
69 |
top = left = right = bottom = 0
|
70 |
-
|
71 |
# expected image variance
|
72 |
hvar,wvar = torch.mean(torch.var(img,dim=0)), torch.mean(torch.var(img,dim=1))
|
73 |
-
|
74 |
# use image variance and alpha to identify too-uniform frame borders
|
75 |
top = torch.var(img[:self.maxPixelFrame,:],dim=1) < wvar*(1+self.alpha)
|
76 |
top = self._map2idx(top)
|
77 |
-
|
78 |
bottom = torch.var(img[-self.maxPixelFrame:,:],dim=1) < wvar*(1+self.alpha)
|
79 |
bottom = self._map2idx(bottom)
|
80 |
-
|
81 |
left = torch.var(img[:,:self.maxPixelFrame],dim=0) < hvar*(1+self.alpha)
|
82 |
left = self._map2idx(left)
|
83 |
-
|
84 |
right = torch.var(img[:,-self.maxPixelFrame:],dim=0) < hvar*(1+self.alpha)
|
85 |
right = self._map2idx(right)
|
86 |
-
|
87 |
return (top,bottom,right,left)
|
88 |
-
|
89 |
def __call__(self,img: Tensor):
|
|
|
90 |
top,bottom,right,left = self._Border(torchvision.transforms.Grayscale()(img)[0])
|
91 |
-
|
92 |
height = img.shape[1]-(top+bottom)
|
93 |
width = img.shape[2]-(left+right)
|
94 |
-
|
95 |
print(f"t{top} b{bottom} l{left} r{right}")
|
96 |
-
|
97 |
-
return torchvision.transforms.functional.crop(img,top,left,height,width)
|
|
|
|
|
|
|
|
|
|
|
1 |
|
2 |
+
import pandas as _pd
|
3 |
+
from skimage import io as _io
|
4 |
+
import matplotlib.pyplot as _plt
|
5 |
+
from pathlib import Path as _Path
|
6 |
+
from PIL import Image as _Image
|
7 |
|
8 |
# Ignore warnings
|
9 |
+
import warnings as _warnings
|
10 |
+
_warnings.filterwarnings("ignore")
|
11 |
|
12 |
+
_plt.ion() # interactive mode
|
13 |
+
|
14 |
+
class MORRIS():
|
15 |
_storage_csv='morris.csv'
|
16 |
_storage_jpg='jpgs'
|
17 |
+
def __init__(self,root=_Path(__file__).parent,transform=False):
|
18 |
+
self.storage = _Path(root)
|
19 |
self.transform = transform
|
20 |
+
self.index = _pd.read_csv(self.storage / self._storage_csv)
|
21 |
+
|
22 |
+
def torch(self):
|
23 |
+
import torch
|
24 |
+
from torchvision import transforms
|
25 |
+
|
26 |
+
class MORRISTORCH(torch.utils.data.Dataset,MORRIS):
|
27 |
+
|
28 |
+
def __init__(self,root=_Path(__file__).parent,transform=False):
|
29 |
+
super().__init__(root,transform)
|
30 |
+
|
31 |
+
def show(self,idx):
|
32 |
+
name=None
|
33 |
+
if isinstance(idx,str):
|
34 |
+
if idx in self.index.name.values:
|
35 |
+
idx = self.index.index[self.index.name==idx][0]
|
36 |
+
idx = int(idx)
|
37 |
+
name = self.index.name[idx]
|
38 |
+
print(f"found item {idx} by name {name}")
|
39 |
+
else:
|
40 |
+
raise ValueError('item name not found')
|
41 |
+
if isinstance(idx,int):
|
42 |
+
name = self.index.name[idx]
|
43 |
+
_plt.title(name)
|
44 |
+
_plt.imshow(transforms.ToPILImage()(self.__getitem__(idx)[0]))
|
45 |
+
else:
|
46 |
+
_plt.imshow(transforms.ToPILImage()(idx))
|
47 |
+
|
48 |
+
def __getitem__(self,idx):
|
49 |
+
item = self.index.iloc[idx].to_dict()
|
50 |
+
image = _io.imread(self.storage / self._storage_jpg / self.index.iloc[idx].filename)
|
51 |
+
image = torch.tensor(image).permute(2,0,1)
|
52 |
+
if self.transform:
|
53 |
+
image = self.transform(image)
|
54 |
+
|
55 |
+
item = [image,item['name'],item['year']]
|
56 |
+
return item
|
57 |
+
|
58 |
+
return MORRISTORCH(str(self.storage),self.transform)
|
59 |
+
|
60 |
def __len__(self):
|
61 |
return len(self.index)
|
62 |
+
|
63 |
def __getitem__(self,idx):
|
64 |
item = self.index.iloc[idx].to_dict()
|
65 |
+
image = _io.imread(self.storage / self._storage_jpg / self.index.iloc[idx].filename)
|
66 |
+
|
67 |
if self.transform:
|
68 |
image = self.transform(image)
|
69 |
+
|
70 |
item['image'] = image
|
71 |
return item
|
72 |
+
|
73 |
+
def show(self,idx):
|
74 |
+
if isinstance(idx,str):
|
75 |
+
if idx in self.index.name.values:
|
76 |
+
idx = self.index.index[self.index.name==idx][0]
|
77 |
+
idx = int(idx)
|
78 |
+
name = self.index.name[idx]
|
79 |
+
print(f"found item {idx} by name {name}")
|
80 |
+
else:
|
81 |
+
raise ValueError('item name not found')
|
82 |
+
if isinstance(idx,int):
|
83 |
+
_item = self.__getitem__(idx)
|
84 |
+
image = _item['image']
|
85 |
+
name = _item['name']
|
86 |
+
_plt.title(name)
|
87 |
+
_plt.imshow(_Image.fromarray(image))
|
88 |
+
else:
|
89 |
+
try:
|
90 |
+
_plt.imshow(_Image.fromarray(idx))
|
91 |
+
except AttributeError:
|
92 |
+
_plt.imshow(_Image.fromarray(idx.permute(1,2,0).numpy()))
|
93 |
+
|
94 |
def _self_validate(self):
|
95 |
"""try loading each image in the dataset"""
|
96 |
allgood=True
|
|
|
101 |
allgood=False
|
102 |
print(f"couldn't load {self.index.iloc[idx].filename}")
|
103 |
if allgood:
|
104 |
+
print(f"All good. {len(self)} images loadable.")
|
105 |
|
106 |
class Deframe(object):
|
107 |
"""check for uniform color boundaries on edges of input and crop them away"""
|
108 |
+
from torch import Tensor
|
109 |
+
|
110 |
def __init__(self,aggressive=False,maxPixelFrame=20):
|
111 |
self.alpha = 0.1 if aggressive else 0.01
|
112 |
self.maxPixelFrame = maxPixelFrame
|
113 |
+
|
114 |
def _map2idx(self,frameMap):
|
115 |
try:
|
116 |
return frameMap.tolist().index(False)
|
117 |
except ValueError:
|
118 |
return self.maxPixelFrame
|
119 |
+
|
120 |
def _Border(self,img: Tensor):
|
121 |
""" take greyscale Tensor
|
122 |
return left,right,top,bottom border size identified """
|
123 |
+
import torch
|
124 |
top = left = right = bottom = 0
|
125 |
+
|
126 |
# expected image variance
|
127 |
hvar,wvar = torch.mean(torch.var(img,dim=0)), torch.mean(torch.var(img,dim=1))
|
128 |
+
|
129 |
# use image variance and alpha to identify too-uniform frame borders
|
130 |
top = torch.var(img[:self.maxPixelFrame,:],dim=1) < wvar*(1+self.alpha)
|
131 |
top = self._map2idx(top)
|
132 |
+
|
133 |
bottom = torch.var(img[-self.maxPixelFrame:,:],dim=1) < wvar*(1+self.alpha)
|
134 |
bottom = self._map2idx(bottom)
|
135 |
+
|
136 |
left = torch.var(img[:,:self.maxPixelFrame],dim=0) < hvar*(1+self.alpha)
|
137 |
left = self._map2idx(left)
|
138 |
+
|
139 |
right = torch.var(img[:,-self.maxPixelFrame:],dim=0) < hvar*(1+self.alpha)
|
140 |
right = self._map2idx(right)
|
141 |
+
|
142 |
return (top,bottom,right,left)
|
143 |
+
|
144 |
def __call__(self,img: Tensor):
|
145 |
+
import torchvision
|
146 |
top,bottom,right,left = self._Border(torchvision.transforms.Grayscale()(img)[0])
|
147 |
+
|
148 |
height = img.shape[1]-(top+bottom)
|
149 |
width = img.shape[2]-(left+right)
|
150 |
+
|
151 |
print(f"t{top} b{bottom} l{left} r{right}")
|
152 |
+
|
153 |
+
return torchvision.transforms.functional.crop(img,top,left,height,width)
|