Detection---RPN

最新推荐文章于 2020-12-18 15:40:18 发布

原创最新推荐文章于 2020-12-18 15:40:18 发布 · 315 阅读

0 ·

CC 4.0 BY-SA版权

object detection 专栏收录该内容

9 篇文章

订阅专栏

本文深入探讨了区域提议网络(RPN)的概念，其作为目标检测的重要组成部分，能够从任意大小的输入图像中生成一系列矩形对象提议，每个提议都附带一个对象得分。文章详细介绍了RPN的架构，包括通过n×n卷积层和两个并行的1×1卷积层实现的区域提议生成过程。此外，还提供了RPN在Python中的代码实现示例，展示了如何生成锚框(anchor)，并将其可视化在输入图像上。

1.定义：

What’s RPN?

A Region Proposal Network(PRN) takes an image (of any size) as input and outputs a set of 
rectangular Object proposals , each with an objectness score.

What’s the architecture of RPN?

This architecture is naturally implemented with an n×n convolutional layer followed
by two sibling 1 × 1 convolutional layers (for reg and cls, respectively).

2.代码实现:

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.patches as patches
import pylab
import cv2

# feature map size
size_Y = 53
size_X = 80
# Scale ratio between input_image to feature map
rpn_stride = 8
# Definition anchor
scales = [1,2,4]
ratios = [0.5,1,2]

def anchor_gen(size_X,size_Y,rpn_stride,scales,ratios):
    scales,ratios = np.meshgrid(scales,ratios)
    scales,ratios = scales.flatten(),ratios.flatten()
    scalesY = scales*np.sqrt(ratios)
    scalesX = scales/np.sqrt(ratios)


    shiftX = np.arange(0, size_X) * rpn_stride
    shiftY = np.arange(0, size_Y) * rpn_stride
    shiftX,shiftY = np.meshgrid(shiftX,shiftY)
    centerX,anchorX = np.meshgrid(shiftX,scalesX)
    centerY,anchorY = np.meshgrid(shiftY,scalesY)
    anchor_center = np.stack([centerY,centerX],axis=2).reshape(-1,2)
    anchor_size = np.stack([anchorY,anchorX],axis=2).reshape(-1,2)
    boxes = np.concatenate([anchor_center-0.5*anchor_size,anchor_center+0.5*anchor_size],axis=1)
    return boxes

if __name__=='__main__':
    anchors = anchor_gen(size_X,size_Y,rpn_stride,scales,ratios)
    plt.figure(figsize=(10,10))
    img = cv2.imread("./person.jpg", 1)
    img = cv2.resize(img,None,fx=0.125,fy=0.125,interpolation=cv2.INTER_CUBIC)
    print(img.shape)
    Axs = plt.gca()

    for i in range(anchors.shape[0]):
        box = anchors[i]
        rec = patches.Rectangle((box[0],box[1]),box[2]-box[0],box[3]-box[1],edgecolor="r",facecolor="none")
        Axs.add_patch(rec)

    plt.imshow(img)
    pylab.show()

3.结果：