Faster-RCNN物体检测---（1）Anchor生成

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
import numpy as np
import keras.utils
import sys
import importlib

import matplotlib.pyplot as plt
import matplotlib.patches as patches
from PIL import Image

#matlotlib inline

#Sample original input
Sample_raw_x = 128
Sample_raw_y = 128

rpn_stride = 8 #8倍下采样

Feature_size_X = Sample_raw_x/rpn_stride
Feature_size_Y = Sample_raw_y/rpn_stride

scales = [1, 2, 4]  #w,h的长
ratios = [0.5, 1, 2] #3的比例

""""
fx = np.arange(Feature_size_X)
fy = np.arange(Feature_size_Y)

FX, FY = np.meshgrid(fx, fy)
FX.flatten()
FY.flatten()

scales,ratios = np.meshgrid(scales,ratios)  #做匹配
scales.flatten()
ratios.flatten()
print(scales.flatten())
print(ratios.flatten())

scale = 4
ratio = 2
x = scale * np.sqrt(ratio)
y = scale / np.sqrt(ratio)
print(x,y) 
"""

def anchor(Feature_size_X, Feature_size_Y, rpn_stride, scales, ratios):
    # 组合尺寸和比例 scales ratios
    scales,ratios = np.meshgrid(scales,ratios)
    scales, ratios = scales.flatten(),ratios.flatten()

    #anchor尺寸计算
    scalesX = scales * np.sqrt(ratios)  #宽度
    scalesY = scales / np.sqrt(ratios)   #长度

    #anchor point 映射关系
    ShiftX =  np.arange(0,Feature_size_X) * rpn_stride
    ShiftY =  np.arange(0,Feature_size_Y) * rpn_stride

    #组合网格  anchor point在原图位置
    ShiftX, ShiftY = np.meshgrid(ShiftX, ShiftY) #XY是Anchor的中心点

    #每个anchor点需要有9个尺寸的anchor框
    centerX, anchorX = np.meshgrid(ShiftX, scalesX)
    centerY, anchorY = np.meshgrid(ShiftY, scalesY)

    #Stack 各种尺寸，各种比例对应各种长度
    anchor_center = np.stack([centerY, centerX], axis=2).reshape(-1, 2) #列对应
    anchor_size = np.stack([anchorY, anchorX], axis=2).reshape(-1, 2)

    #左上右下坐标点输出  #1维上x的内容
    boxes = np.concatenate([anchor_center - 0.5 * anchor_size,anchor_center + 0.5 * anchor_size],axis=1)
    return boxes
anchors = anchor(Feature_size_X, Feature_size_Y, rpn_stride, scales, ratios)
"""
print(anchors.shape)
print(anchors[1000])
"""
plt.figure(figsize=(10,10))
image = Image.open('test.jpg') #128 * 128
plt.imshow(image)
#plt.show()
asx = plt.gca()  #get current axs
for i in range(anchors.shape[0]):
    box = anchors[i]
    #  print(box)
    rec = patches.Rectangle((box[0], box[1]), box[2]-box[0], box[3]-box[1], edgecolor='r', facecolor='none') #打印小片,红色不填充
    asx.add_patch(rec)
plt.show()