-
Notifications
You must be signed in to change notification settings - Fork 1
/
img_patch.py
92 lines (74 loc) · 2.95 KB
/
img_patch.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
import numpy as np
from collections import Iterable
import math
import img_utils as wmli
class ImagePatch:
def __init__(self,patch_size,pad=True,pad_value=127,boundary=0) -> None:
'''
patch_size: (H,W)
boundary: (bh,bw) or value
'''
self.patch_size = patch_size
self.pad = pad
self.pad_value = pad_value
self.boundary = boundary if isinstance(boundary,Iterable) else (boundary,boundary)
self.patch_bboxes = []
self.src_img = None
self.cur_idx = 0
def set_src_img(self,img):
self.src_img = img
self.cur_idx = 0
self.patch_bboxes = []
self.rows = math.ceil((self.src_img.shape[0]-self.boundary[0])/(self.patch_size[0]-self.boundary[0]))
self.cols = math.ceil((self.src_img.shape[1]-self.boundary[1])/(self.patch_size[1]-self.boundary[1]))
x = np.array(list(range(self.cols)),dtype=np.int32)*(self.patch_size[1]-self.boundary[1])
y = np.array(list(range(self.rows)),dtype=np.int32)*(self.patch_size[0]-self.boundary[0])
wh = np.array([self.patch_size[1],self.patch_size[0]],dtype=np.int32)
wh = np.reshape(wh,[-1,2])
xv,yv = np.meshgrid(x,y,sparse=False, indexing='ij')
x0y0 = np.stack([xv,yv],axis=-1)
x0y0 = np.reshape(x0y0,[-1,x0y0.shape[-1]])
x1y1 = x0y0+wh
self.bboxes = np.concatenate([x0y0,x1y1],axis=-1)
def __len__(self):
return len(self.bboxes)
def __getitem__(self,idx):
bbox = self.bboxes[idx]
self.cur_idx = idx
if self.pad:
size = self.patch_size[::-1]
return wmli.crop_and_pad(self.src_img,bbox,size,pad_color=self.pad_value)
else:
return wmli.crop_img(self.src_img,bbox)
def patch_bboxes2img_bboxes(self,bboxes,idx=None):
'''
bboxes: [N,4] (x0,y0,x1,y1)
'''
if idx is None:
idx = self.cur_idx
bbox = self.bboxes[idx]
offset = np.array([bbox[0],bbox[1],bbox[0],bbox[1]],dtype=bboxes.dtype)
offset = np.reshape(offset,[-1,4])
bboxes = bboxes+offset
return bboxes
def cur_bbox(self):
return self.bboxes[self.cur_idx]
def remove_boundary_bboxes(self,bboxes,boundary=None):
'''
bboxes: [N,4] (x0,y0,x1,y1), in patch img
'''
if boundary is None:
boundary = self.boundary
if not isinstance(boundary,Iterable):
boundary = (boundary,boundary)
value = (boundary[0]/2,boundary[1]/2)
cxy = (bboxes[...,:2]+bboxes[...,2:])/2
mask0 = cxy[...,0]<value[1]
mask1 = cxy[...,1]<value[0]
mask2 = cxy[...,0]>(self.patch_size[1]-value[1])
mask3 = cxy[...,1]>(self.patch_size[0]-value[0])
_mask0 = np.logical_or(mask0,mask1)
_mask2 = np.logical_or(mask2,mask3)
mask = np.logical_or(_mask0,_mask2)
keep = np.logical_not(mask)
return keep