Faster Rcnn 源码解析（一）—— anchor_targte_layer.py

最新推荐文章于 2021-08-19 14:24:18 发布

原创最新推荐文章于 2021-08-19 14:24:18 发布 · 844 阅读

3 ·

CC 4.0 BY-SA版权

文章标签：

#深度学习 #源码 #FasterRcnn #anchor_targte_layer.py

深度学习专栏收录该内容

10 篇文章

订阅专栏

本文解析了Faster R-CNN中的AnchorTargetLayer工作原理，包括如何生成锚框、确定正负样本及其权重等关键步骤。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

AnchorTargetLayer层

功能：

得到所有的anchor，根据GT确定每个anchor的标签，并得到anchor与最大IOU的GT的偏移量
个人理解：这里就相当于是得到了每个anchor要学习的目标。

输入：

bottom: 'rpn_cls_score'#只是为了确定当前feature map的height、width
bottom: 'gt_boxes'框的ground truth[x,y,w,h]
bottom: 'im_info'图片的大小和当前的尺度
bottom: 'data'输入的图片的信息

输出：

top: 'rpn_labels'大小是[1，1，A*height，width]，A是anchar的数目
top: 'rpn_bbox_targets'大小是[1，A*4，height，width]：anchor和最高重叠gt的偏移量
top: 'rpn_bbox_inside_weights'大小是[1，A*4，height，width]：被抽中的正类为1，其他为0。在做回归的时候只对前景做
top: 'rpn_bbox_outside_weights'大小是[1，A*4，height，width]：外部权重，目前负例的外部权重=正例的外部权重=1/Nreg

流程：

（1）根据feature map的大小和_feat_stride得到all_anchors,大小是（K*A），这里feat_stride=16，可以理解为rpn_cls_score映射到原图的坐标点，K=height*width。
 (2)过滤不在图片内部的得到anchors。
 (3)计算anchors和gt_boxes的overlap，判断K*A个那些为正，那些为负。
 (4)最后labels中存在的是抽样的，抽128个fg，正样本不够128，负样本多取点，凑够256个。正样本=1，负样本=0，不用的赋值为-1。
 (5)计算rpn_bbox_targets，rpn_bbox_inside_weights，rpn_bbox_outside_weights。

源码：

# --------------------------------------------------------
# Faster R-CNN
# Copyright (c) 2015 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Ross Girshick and Sean Bell
# --------------------------------------------------------

import os
import caffe
import yaml
from fast_rcnn.config import cfg
import numpy as np
import numpy.random as npr
from generate_anchors import generate_anchors
from utils.cython_bbox import bbox_overlaps
from fast_rcnn.bbox_transform import bbox_transform

DEBUG = False

class AnchorTargetLayer(caffe.Layer):
    """
    Assign anchors to ground-truth targets. Produces anchor classification
    labels and bounding-box regression targets.
    """

    def setup(self, bottom, top):
        layer_params = yaml.load(self.param_str_)
        anchor_scales = layer_params.get('scales', (8, 16, 32))
        self._anchors = generate_anchors(scales=np.array(anchor_scales))#相关代码可以参考这个博客
        self._num_anchors = self._anchors.shape[0]# 9个
        self._feat_stride = layer_params['feat_stride']#对应于generate_anchors中的base_size是16

        if DEBUG:
            print 'anchors:'
            print self._anchors
            print 'anchor shapes:'#打印每个anchor的形状（width,height）
            print np.hstack((
                self._anchors[:, 2::4] - self._anchors[:, 0::4],
                self._anchors[:, 3::4] - self._anchors[:, 1::4],
            ))
            self._counts = cfg.EPS #config.py 里面的一个参数
            self._sums = np.zeros((1, 4))
            self._squared_sums = np.zeros((1, 4))
            self._fg_sum = 0
            self._bg_sum = 0
            self._count = 0

        # allow boxes to sit over the edge by a small amount
        self._allowed_border = layer_params.get('allowed_border', 0)

        height, width = bottom[0].data.shape[-2:]
        if DEBUG:
            print 'AnchorTargetLayer: height', height, 'width', width

        A = self._num_anchors #一般设置为9.
        # 定义输出
        # 在这里将top的维度结构reshape
        # labels
        top[0].reshape(1, 1, A * height, width)
        # bbox_targets
        top[1].reshape(1, A * 4, height, width)
        # bbox_inside_weights
        top[2].reshape(1, A * 4, height, width)
        # bbox_outside_weights
        top[3].reshape(1, A * 4, height, width)

    def forward(self, bottom, top):
        # Algorithm:
        #
        # for each (H, W) location i
        #   generate 9 anchor boxes centered on cell i
        #   apply predicted bbox deltas at cell i to each of the 9 anchors
        # filter out-of-image anchors
        # measure GT overlap

        assert bottom[0].data.shape[0] == 1, \
            'Only single item batches are supported'

        # feature map of shape (..., H, W),特征图的大小
        height, width = bottom[0].data.shape[-2:]
        # GT boxes (x1, y1, x2, y2, label)
        gt_boxes = bottom[1].data
        # im_info
        im_info = bottom[2].data[0, :]

        if DEBUG:
            print ''
            print 'im_size: ({}, {})'.format(im_info[0], im_info[1])
            print 'scale: {}'.format(im_info[2])
            print 'height, width: ({}, {})'.format(height, width)
            print 'rpn: gt_boxes.shape', gt_boxes.shape
            print 'rpn: gt_boxes', gt_boxes

        # 1. Generate proposals from bbox deltas and shifted anchors
        shift_x = np.arange(0, width) * self._feat_stride #x方向的偏移量大小
        shift_y = np.arange(0, height) * self._feat_stride #y方向的偏移量大小
        #以_feat_stride = 16为例 ：
        # shift_x =（0, 16, 32，...，width*_feat_stride），
        #shift_y =（0, 16, 32，...，height*_feat_stride），

        # shift_x，shift_y均为width×height的二维数组，若width*height = 39×64
        shift_x, shift_y = np.meshgrid(shift_x, shift_y)
        # 对应位置的元素组合即构成图像上需要偏移量大小（偏移量大小是相对与图像最左上角的那9个anchor的偏移量大小）
        # 也就是说总共会得到2496个偏移值对。这些偏移值对与初始的anchor相加即可得到所有的anchors，
        # 总共会产生2496×9个anchors，且存储在all_anchors变量中
        shifts = np.vstack((shift_x.ravel(), shift_y.ravel(),
                            shift_x.ravel(), shift_y.ravel())).transpose()
        # add A anchors (1, A, 4) to
        # cell K shifts (K, 1, 4) to get
        # shift anchors (K, A, 4)
        # reshape to (K*A, 4) shifted anchors
        A = self._num_anchors
        K = shifts.shape[0]
        all_anchors = (self._anchors.reshape((1, A, 4)) +
                       shifts.reshape((1, K, 4)).transpose((1, 0, 2)))
        all_anchors = all_anchors.reshape((K * A, 4))
        total_anchors = int(K * A)

        # only keep anchors inside the image
        inds_inside = np.where(
            (all_anchors[:, 0] >= -self._allowed_border) &
            (all_anchors[:, 1] >= -self._allowed_border) &
            (all_anchors[:, 2] < im_info[1] + self._allowed_border) &  # width
            (all_anchors[:, 3] < im_info[0] + self._allowed_border)    # height
        )[0]

        if DEBUG:
            print 'total_anchors', total_anchors
            print 'inds_inside', len(inds_inside)

        # keep only inside anchors
        #得到所有在图像边界内部anchors
        anchors = all_anchors[inds_inside, :]
        if DEBUG:
            print 'anchors.shape', anchors.shape

        # label: 1 is positive, 0 is negative, -1 is dont care
        #产生与anchors对应大小的label，初始化为-1.
        labels = np.empty((len(inds_inside), ), dtype=np.float32)
        labels.fill(-1)

        # overlaps between the anchors and the gt boxes
        # overlaps (ex, gt)
        # 这里overlaps是计算所有anchor与ground-truth的重合度，
        # 它是一个len(anchors) x len(gt_boxes)的二维数组，每个元素是各个anchor和gt_boxes的overlap值
        # overlap = (重合部分面积) / (anchor面积 + gt_boxes面积 - 重合部分面积)
        overlaps = bbox_overlaps(
            np.ascontiguousarray(anchors, dtype=np.float),#返回一个连续的浮点型数组。
            np.ascontiguousarray(gt_boxes, dtype=np.float))

        # argmax_overlaps是每个anchor对应最大overlap的gt_boxes的下标,返回的是每一行的最大值，行向量
        # max_overlaps是每个anchor对应最大的overlap值
        argmax_overlaps = overlaps.argmax(axis=1)
        max_overlaps = overlaps[np.arange(len(inds_inside)), argmax_overlaps]

        # gt_argmax_overlaps是每个gt_boxes对应最大overlap的anchor的下标,返回的是每一列的最大值，行向量
        # gt_max_overlaps是每个gt_boxes对应最大的overlap值
        gt_argmax_overlaps = overlaps.argmax(axis=0)
        gt_max_overlaps = overlaps[gt_argmax_overlaps,
                                   np.arange(overlaps.shape[1])]

        # 加上这一步是因为有很多overlap并列第一，要把所有的都找出来
        gt_argmax_overlaps = np.where(overlaps == gt_max_overlaps)[0]
        #接下来就是根据overlap的值确定每个anchor是前景还是背景

        # RPN_CLOBBER_POSITIVES = false，先按照RPN_NEGATIVE_OVERLAP挑选bg，这样bg可能变成fg；
        # RPN_CLOBBER_POSITIVES = true，最后挑选bg，这样fg可能变成bg；
        if not cfg.TRAIN.RPN_CLOBBER_POSITIVES:
            # assign bg labels first so that positive labels can clobber them
            labels[max_overlaps < cfg.TRAIN.RPN_NEGATIVE_OVERLAP] = 0

        # fg label: for each gt, anchor with highest overlap
        # 对于某个gt，overlap最大的anchor为1
        labels[gt_argmax_overlaps] = 1

        # fg label: above threshold IOU
        # 对于某个anchor，其overlap超过阈值为1
        labels[max_overlaps >= cfg.TRAIN.RPN_POSITIVE_OVERLAP] = 1

        if cfg.TRAIN.RPN_CLOBBER_POSITIVES:
            # assign bg labels last so that negative labels can clobber positives
            labels[max_overlaps < cfg.TRAIN.RPN_NEGATIVE_OVERLAP] = 0

        # subsample positive labels if we have too many
        # 接下来是确定正负样本的数量
        # RPN_FG_FRACTION：rpn样本数中，fg的比例 RPN_BATCHSIZE：rpn样本数
        num_fg = int(cfg.TRAIN.RPN_FG_FRACTION * cfg.TRAIN.RPN_BATCHSIZE)#训练需要的正样本的个数
        fg_inds = np.where(labels == 1)[0]#所有的正样本的个数
        #如果正样本的数量太多，随机挑选一本份多余的置为-1（无效）.
        if len(fg_inds) > num_fg:
            disable_inds = npr.choice(
                fg_inds, size=(len(fg_inds) - num_fg), replace=False)
            labels[disable_inds] = -1

        # subsample negative labels if we have too many
        num_bg = cfg.TRAIN.RPN_BATCHSIZE - np.sum(labels == 1)#训练需要的负样本的个数
        bg_inds = np.where(labels == 0)[0]#所有的负样本的个数
        # 如果正样本的数量太多，随机挑选一本份多余的置为-1（无效）.
        if len(bg_inds) > num_bg:
            disable_inds = npr.choice(
                bg_inds, size=(len(bg_inds) - num_bg), replace=False)
            labels[disable_inds] = -1
            #print "was %s inds, disabling %s, now %s inds" % (
                #len(bg_inds), len(disable_inds), np.sum(labels == 0))
        # 这里将计算每一个anchor与重合度最高的ground_truth的偏移值，
        # 详细的计算方法在论文中提到，
        # 在fast-rcnn/bbox_transform.py中的bbox_transform函数也非常容易看懂
        bbox_targets = np.zeros((len(inds_inside), 4), dtype=np.float32)
        bbox_targets = _compute_targets(anchors, gt_boxes[argmax_overlaps, :])
        # bbox_inside_weights的含义是只计算前景的回归，
        # 所以他的定义就是除了前景为（1, 1, 1, 1），其余的都是（0，0，0，0）
        bbox_inside_weights = np.zeros((len(inds_inside), 4), dtype=np.float32)
        bbox_inside_weights[labels == 1, :] = np.array(cfg.TRAIN.RPN_BBOX_INSIDE_WEIGHTS)
        #bbox_outside_weights是为了在函数中加入前景和背景的权重，这里权重相同，都为使用的anchor的数量。
        bbox_outside_weights = np.zeros((len(inds_inside), 4), dtype=np.float32)
        if cfg.TRAIN.RPN_POSITIVE_WEIGHT < 0: #正负样本具有一样的权重
            # uniform weighting of examples (given non-uniform sampling)
            num_examples = np.sum(labels >= 0) #所有样本的数量
            positive_weights = np.ones((1, 4)) * 1.0 / num_examples
            negative_weights = np.ones((1, 4)) * 1.0 / num_examples
        else:
            assert ((cfg.TRAIN.RPN_POSITIVE_WEIGHT > 0) & #必须在0-1，否则会抛出异常
                    (cfg.TRAIN.RPN_POSITIVE_WEIGHT < 1))
            positive_weights = (cfg.TRAIN.RPN_POSITIVE_WEIGHT /
                                np.sum(labels == 1))
            negative_weights = ((1.0 - cfg.TRAIN.RPN_POSITIVE_WEIGHT) /
                                np.sum(labels == 0))
        bbox_outside_weights[labels == 1, :] = positive_weights
        bbox_outside_weights[labels == 0, :] = negative_weights

        if DEBUG:
            #计算正样本的偏移量的均值和方差
            self._sums += bbox_targets[labels == 1, :].sum(axis=0)
            self._squared_sums += (bbox_targets[labels == 1, :] ** 2).sum(axis=0)
            self._counts += np.sum(labels == 1)
            means = self._sums / self._counts
            stds = np.sqrt(self._squared_sums / self._counts - means ** 2)
            print 'means:'
            print means
            print 'stdevs:'
            print stds

        # map up to original set of anchors
        # 还记得文初将all_anchors裁减掉了2/3左右，仅仅保留在图像内的anchor吗，
        # 这里就是将其复原作为下一层的输入了
        labels = _unmap(labels, total_anchors, inds_inside, fill=-1)
        bbox_targets = _unmap(bbox_targets, total_anchors, inds_inside, fill=0)
        bbox_inside_weights = _unmap(bbox_inside_weights, total_anchors, inds_inside, fill=0)
        bbox_outside_weights = _unmap(bbox_outside_weights, total_anchors, inds_inside, fill=0)

        if DEBUG:
            print 'rpn: max max_overlap', np.max(max_overlaps)
            print 'rpn: num_positive', np.sum(labels == 1)
            print 'rpn: num_negative', np.sum(labels == 0)
            self._fg_sum += np.sum(labels == 1)
            self._bg_sum += np.sum(labels == 0)
            self._count += 1
            print 'rpn: num_positive avg', self._fg_sum / self._count
            print 'rpn: num_negative avg', self._bg_sum / self._count
        #将输出reshape成相应的格式。
        # labels
        labels = labels.reshape((1, height, width, A)).transpose(0, 3, 1, 2)
        labels = labels.reshape((1, 1, A * height, width))
        top[0].reshape(*labels.shape)
        top[0].data[...] = labels

        # bbox_targets
        bbox_targets = bbox_targets \
            .reshape((1, height, width, A * 4)).transpose(0, 3, 1, 2)
        top[1].reshape(*bbox_targets.shape)
        top[1].data[...] = bbox_targets

        # bbox_inside_weights
        bbox_inside_weights = bbox_inside_weights \
            .reshape((1, height, width, A * 4)).transpose(0, 3, 1, 2)
        assert bbox_inside_weights.shape[2] == height
        assert bbox_inside_weights.shape[3] == width
        top[2].reshape(*bbox_inside_weights.shape)
        top[2].data[...] = bbox_inside_weights

        # bbox_outside_weights
        bbox_outside_weights = bbox_outside_weights \
            .reshape((1, height, width, A * 4)).transpose(0, 3, 1, 2)
        assert bbox_outside_weights.shape[2] == height
        assert bbox_outside_weights.shape[3] == width
        top[3].reshape(*bbox_outside_weights.shape)
        top[3].data[...] = bbox_outside_weights

    def backward(self, top, propagate_down, bottom):
        """This layer does not propagate gradients."""
        pass

    def reshape(self, bottom, top):
        """Reshaping happens during the call to forward."""
        pass

# 输入有两种：一维的labels,e二维的bbox_targets，bbox_inside_weights，bbox_outside_weights
# labels = _unmap(labels, total_anchors, inds_inside, fill=-1)
# bbox_targets = _unmap(bbox_targets, total_anchors, inds_inside, fill=0)
# bbox_inside_weights = _unmap(bbox_inside_weights, total_anchors, inds_inside, fill=0)
# bbox_outside_weights = _unmap(bbox_outside_weights, total_anchors, inds_inside, fill=0)
def _unmap(data, count, inds, fill=0):
    """ Unmap a subset of item (data) back to the original set of items (of
    size count) """
    if len(data.shape) == 1:
        ret = np.empty((count, ), dtype=np.float32)
        ret.fill(fill)
        ret[inds] = data
    else:
        ret = np.empty((count, ) + data.shape[1:], dtype=np.float32)
        ret.fill(fill)
        ret[inds, :] = data
    return ret


def _compute_targets(ex_rois, gt_rois):
    """Compute bounding-box regression targets for an image."""

    assert ex_rois.shape[0] == gt_rois.shape[0]
    assert ex_rois.shape[1] == 4
    assert gt_rois.shape[1] == 5

    return bbox_transform(ex_rois, gt_rois[:, :4]).astype(np.float32, copy=False)#bbox_transform请戳