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