Faster RCNN minibatch.py解读
来源:互联网 发布:spss数据显著性分析 编辑:程序博客网 时间:2024/06/01 09:57
minibatch.py 的功能是: Compute minibatch blobs for training a Fast R-CNN network. 与roidb不同的是, minibatch中存储的并不是完整的整张图像图像,而是从图像经过转换后得到的四维blob以及从图像中截取的proposals,以及与之对应的labels等
在整个faster rcnn训练中,有两处用到了minibatch.py,一处是rpn的开始数据输入,另一处自然是fast rcnn的数据输入。分别见stage1_rpn_train.pt与stage1_fast_rcnn_train.py的最前面,如下:
stage1_rpn_train.pt:
layer { name: 'input-data' type: 'Python' top: 'data' top: 'im_info' top: 'gt_boxes' python_param { module: 'roi_data_layer.layer' layer: 'RoIDataLayer' param_str: "'num_classes': 21" }}
stage1_fast_rcnn_train.py:
name: "VGG_CNN_M_1024"layer { name: 'data' type: 'Python' top: 'data' top: 'rois' top: 'labels' top: 'bbox_targets' top: 'bbox_inside_weights' top: 'bbox_outside_weights' python_param { module: 'roi_data_layer.layer' layer: 'RoIDataLayer' param_str: "'num_classes': 21" }}
如上,共同的数据定义层为roi_data_layer.layer,在layer.py中,观察前向传播:
def forward(self, bottom, top): """Get blobs and copy them into this layer's top blob vector.""" blobs = self._get_next_minibatch() for blob_name, blob in blobs.iteritems(): top_ind = self._name_to_top_map[blob_name] # Reshape net's input blobs top[top_ind].reshape(*(blob.shape)) # Copy data into net's input blobs top[top_ind].data[...] = blob.astype(np.float32, copy=False) def _get_next_minibatch(self): """Return the blobs to be used for the next minibatch. If cfg.TRAIN.USE_PREFETCH is True, then blobs will be computed in a separate process and made available through self._blob_queue. """ if cfg.TRAIN.USE_PREFETCH: return self._blob_queue.get() else: db_inds = self._get_next_minibatch_inds() minibatch_db = [self._roidb[i] for i in db_inds] return get_minibatch(minibatch_db, self._num_classes)
这时我们发现了get_minibatch,此函数出现在minibatch.py中。
在看这份代码的时候,建议从get_minibatch开始。下面我们开始:
get_minibatch中,【输入】:roidb是一个list,list中的每个元素是一个字典,每个字典对应一张图片的信息,其中的主要信息有:
num_classes在pascal_voc中为21.
def get_minibatch(roidb, num_classes): """Given a roidb, construct a minibatch sampled from it.""" # 给定一个roidb,这个roidb中存储的可能是多张图片,也可能是单张或者多张图片, num_images = len(roidb) # Sample random scales to use for each image in this batch random_scale_inds = npr.randint(0, high=len(cfg.TRAIN.SCALES), size=num_images) assert(cfg.TRAIN.BATCH_SIZE % num_images == 0), \ 'num_images ({}) must divide BATCH_SIZE ({})'. \ format(num_images, cfg.TRAIN.BATCH_SIZE) rois_per_image = cfg.TRAIN.BATCH_SIZE / num_images #这里在fast rcnn中,为128/2=64 fg_rois_per_image = np.round(cfg.TRAIN.FG_FRACTION * rois_per_image)#这里比例为0.25=1/4 # Get the input image blob, formatted for caffe #将给定的roidb经过预处理(resize以及resize的scale), #然后再利用im_list_to_blob函数来将图像转换成caffe支持的数据结构,即 N * C * H * W的四维结构 im_blob, im_scales = _get_image_blob(roidb, random_scale_inds) blobs = {'data': im_blob} if cfg.TRAIN.HAS_RPN:#用在rpn assert len(im_scales) == 1, "Single batch only" assert len(roidb) == 1, "Single batch only" # gt boxes: (x1, y1, x2, y2, cls) gt_inds = np.where(roidb[0]['gt_classes'] != 0)[0] gt_boxes = np.empty((len(gt_inds), 5), dtype=np.float32) gt_boxes[:, 0:4] = roidb[0]['boxes'][gt_inds, :] * im_scales[0] gt_boxes[:, 4] = roidb[0]['gt_classes'][gt_inds] blobs['gt_boxes'] = gt_boxes #首先解释im_info。对于一副任意大小PxQ图像,传入Faster RCNN前首先reshape到固定MxN,im_info=[M, N, scale_factor]则保存了此次缩放的所有信息。 blobs['im_info'] = np.array( [[im_blob.shape[2], im_blob.shape[3], im_scales[0]]], dtype=np.float32) else: # not using RPN ,用在fast rcnn # Now, build the region of interest and label blobs rois_blob = np.zeros((0, 5), dtype=np.float32) labels_blob = np.zeros((0), dtype=np.float32) bbox_targets_blob = np.zeros((0, 4 * num_classes), dtype=np.float32) bbox_inside_blob = np.zeros(bbox_targets_blob.shape, dtype=np.float32) # all_overlaps = [] for im_i in xrange(num_images): # 遍历给定的roidb中的每张图片,随机组合sample of RoIs, 来生成前景样本和背景样本。 # 返回包括每张图片中的roi(proposal)的坐标,所属的类别,bbox回归目标,bbox的inside_weight等 labels, overlaps, im_rois, bbox_targets, bbox_inside_weights \ = _sample_rois(roidb[im_i], fg_rois_per_image, rois_per_image, num_classes) # Add to RoIs blob # _sample_rois返回的im_rois并没有缩放,所以这里要先缩放 rois = _project_im_rois(im_rois, im_scales[im_i]) batch_ind = im_i * np.ones((rois.shape[0], 1)) rois_blob_this_image = np.hstack((batch_ind, rois))# 加上图片的序号,共5列(index,x1,y1,x2,y2) rois_blob = np.vstack((rois_blob, rois_blob_this_image)) # 将所有的盒子竖着摆放,如下: # n x1 y1 x2 y2 # 0 .. .. .. .. # 0 .. .. .. .. # : : : : : # 1 .. .. .. .. # 1 .. .. .. .. # Add to labels, bbox targets, and bbox loss blobs labels_blob = np.hstack((labels_blob, labels))# 水平向量,一维向量 bbox_targets_blob = np.vstack((bbox_targets_blob, bbox_targets)) bbox_inside_blob = np.vstack((bbox_inside_blob, bbox_inside_weights)) # 将所有的bbox_targets_blob竖着摆放,如下: N*4k ,只有对应的类非0 # tx1 ty1 wx1 wy1 tx2 ty2 wx2 wy2 tx3 ty3 wx3 wy3 # 0 0 0 0 0 0 0 0 0 0 0 0 # 0 0 0 0 0.2 0.3 1.0 0.5 0 0 0 0 # 0 0 0 0 0 0 0 0 0 0 0 0 # 0 0 0 0 0 0 0 0 0.5 0.5 1.0 1.0 # 0 0 0 0 0 0 0 0 0 0 0 0 # 对于bbox_inside_blob ,与bbox_targets_blob 规模相同,只不过把上面非0的元素换成1即可。 # all_overlaps = np.hstack((all_overlaps, overlaps)) # For debug visualizations # _vis_minibatch(im_blob, rois_blob, labels_blob, all_overlaps) blobs['rois'] = rois_blob blobs['labels'] = labels_blob if cfg.TRAIN.BBOX_REG: blobs['bbox_targets'] = bbox_targets_blob blobs['bbox_inside_weights'] = bbox_inside_blob blobs['bbox_outside_weights'] = \ np.array(bbox_inside_blob > 0).astype(np.float32)#对于bbox_outside_weights,此处看来与bbox_inside_blob 相同。 return blobs
在 def get_minibatch(roidb, num_classes) 中调用此函数,传进来的实参为单张图像的roidb ,该函数主要功能是随机组合sample of RoIs, 来生成前景样本和背景样本。这里很重要,
因为一般来说,生成的proposal背景类比较多,所以我们生成前景与背景的比例选择为1:3,所以
这里每张图片选取了1/4*64=16个前景,选取了3/4*64=48个背景box.
还有一个值得注意的是随机采样中,前景box可能会包含ground truth box.可能会参与分类,但是不会参加回归,因为其回归量为0. 是不是可以将
fg_inds = np.where(overlaps >= cfg.TRAIN.FG_THRESH)[0]
改为:
fg_inds = np.where(overlaps >= cfg.TRAIN.FG_THRESH && overlaps <1.0)[0]
会更合适呢,这样就可以提取的全部是rpn的 proposal。
def _sample_rois(roidb, fg_rois_per_image, rois_per_image, num_classes): """Generate a random sample of RoIs comprising foreground and background examples. """ # label = class RoI has max overlap with labels = roidb['max_classes'] overlaps = roidb['max_overlaps'] rois = roidb['boxes'] # Select foreground RoIs as those with >= FG_THRESH overlap fg_inds = np.where(overlaps >= cfg.TRAIN.FG_THRESH)[0] # Guard against the case when an image has fewer than fg_rois_per_image # foreground RoIs fg_rois_per_this_image = np.minimum(fg_rois_per_image, fg_inds.size) # Sample foreground regions without replacement if fg_inds.size > 0: fg_inds = npr.choice( fg_inds, size=fg_rois_per_this_image, replace=False) # Select background RoIs as those within [BG_THRESH_LO, BG_THRESH_HI) bg_inds = np.where((overlaps < cfg.TRAIN.BG_THRESH_HI) & (overlaps >= cfg.TRAIN.BG_THRESH_LO))[0] # Compute number of background RoIs to take from this image (guarding # against there being fewer than desired) bg_rois_per_this_image = rois_per_image - fg_rois_per_this_image bg_rois_per_this_image = np.minimum(bg_rois_per_this_image, bg_inds.size) # Sample foreground regions without replacement if bg_inds.size > 0: bg_inds = npr.choice( bg_inds, size=bg_rois_per_this_image, replace=False) # The indices that we're selecting (both fg and bg) keep_inds = np.append(fg_inds, bg_inds) # Select sampled values from various arrays: labels = labels[keep_inds] # Clamp labels for the background RoIs to 0 labels[fg_rois_per_this_image:] = 0 overlaps = overlaps[keep_inds] rois = rois[keep_inds] # 调用_get_bbox_regression_labels函数,生成bbox_targets 和 bbox_inside_weights, #它们都是N * 4K 的ndarray,N表示keep_inds的size,也就是minibatch中样本的个数;bbox_inside_weights #也随之生成 bbox_targets, bbox_inside_weights = _get_bbox_regression_labels( roidb['bbox_targets'][keep_inds, :], num_classes) return labels, overlaps, rois, bbox_targets, bbox_inside_weights
def _get_bbox_regression_labels(bbox_target_data, num_classes):
该函数主要是获取bbox_target_data中回归目标的的4个坐标编码作为bbox_targets,同时生成bbox_inside_weights,它们都是N * 4K 的ndarray,N表示keep_inds的size,也就是minibatch中样本的个数。
bbox_target_data: N*5 ,每一行为(c,tx,ty,tw,th)
def _get_bbox_regression_labels(bbox_target_data, num_classes): """Bounding-box regression targets are stored in a compact form in the roidb. This function expands those targets into the 4-of-4*K representation used by the network (i.e. only one class has non-zero targets). The loss weights are similarly expanded. Returns: bbox_target_data (ndarray): N x 4K blob of regression targets bbox_inside_weights (ndarray): N x 4K blob of loss weights """ clss = bbox_target_data[:, 0] bbox_targets = np.zeros((clss.size, 4 * num_classes), dtype=np.float32) bbox_inside_weights = np.zeros(bbox_targets.shape, dtype=np.float32) inds = np.where(clss > 0)[0] # 取前景框 for ind in inds: cls = clss[ind] start = 4 * cls end = start + 4 bbox_targets[ind, start:end] = bbox_target_data[ind, 1:] bbox_inside_weights[ind, start:end] = cfg.TRAIN.BBOX_INSIDE_WEIGHTS return bbox_targets, bbox_inside_weights
对于roidb的图像进行对应的缩放操作,并返回统一的blob数据,即 N * C * H * W(这里为2*3*600*1000)的四维结构
def _get_image_blob(roidb, scale_inds): """Builds an input blob from the images in the roidb at the specified scales. """ num_images = len(roidb) processed_ims = [] im_scales = [] for i in xrange(num_images): im = cv2.imread(roidb[i]['image']) #shape:h*w*c if roidb[i]['flipped']: im = im[:, ::-1, :] # 水平翻转 target_size = cfg.TRAIN.SCALES[scale_inds[i]] im, im_scale = prep_im_for_blob(im, cfg.PIXEL_MEANS, target_size, cfg.TRAIN.MAX_SIZE) im_scales.append(im_scale) processed_ims.append(im) # Create a blob to hold the input images blob = im_list_to_blob(processed_ims) return blob, im_scales
以上im_list_to_blob中将一系列的图像转化为标准的4维矩阵,进行了填0的补全操作,使得所有的图片的大小相同。
prep_im_for_blob 进行尺寸变化,使得最小的边长为target_size,最大的边长不超过cfg.TRAIN.MAX_SIZE,并且返回缩放的比例。
def prep_im_for_blob(im, pixel_means, target_size, max_size): """Mean subtract and scale an image for use in a blob.""" im = im.astype(np.float32, copy=False) im -= pixel_means im_shape = im.shape im_size_min = np.min(im_shape[0:2]) im_size_max = np.max(im_shape[0:2]) im_scale = float(target_size) / float(im_size_min) # Prevent the biggest axis from being more than MAX_SIZE if np.round(im_scale * im_size_max) > max_size: im_scale = float(max_size) / float(im_size_max) im = cv2.resize(im, None, None, fx=im_scale, fy=im_scale, interpolation=cv2.INTER_LINEAR) return im, im_scale
所以对于原始的图片,要缩放到标准的roidb的data的格式,实际上只需要乘以im_scale即可。
反之,如果回到原始的图片,则只需要除以im_scale即可。
参考文献
- http://blog.csdn.net/iamzhangzhuping/article/details/51393032
- faster-rcnn 之 基于roidb get_minibatch(数据准备操作)
- faster rcnn源码解读(六)之minibatch
- Faster RCNN minibatch.py解读
- Faster RCNN minibatch.py
- faster rcnn源码解读(六)之minibatch
- faster rcnn源码解读(六)之minibatch
- Faster-RCNN_TF代码解读15:roi_data_layer/minibatch.py
- py-faster-rcnn源码解读系列
- py-faster-rcnn源码解读系列
- 【py-faster-rcnn】各函数作用解读
- py-faster-rcnn测试流程解读
- py-faster-rcnn代码roidb.py的解读
- faster rcnn源码解读(三)train_faster_rcnn_alt_opt.py
- faster rcnn源码解读(三)train_faster_rcnn_alt_opt.py
- Faster RCNN roidb.py
- Faster RCNN train_faster_rcnn_alt_opt.py
- Faster RCNN layer.py
- Faster RCNN train.py
- Faster RCNN generate.py
- Faster RCNN blob.py
- SpringMVC 实现文件的上传与下载--较为原始
- Java设计模式--构建者模式(Builder)
- 至此,再没有宏伟壮志,但求安稳
- Set集合HashSet学习(Set集合学习一)
- jsPlumb流程图
- Faster RCNN minibatch.py解读
- 第九章 排序
- 从零开始搞懂Android网络请求——HttpURLConnection
- 51nod 1050 循环数组最大子段和
- 11.06 CSS的基础应用2
- 卡尔曼滤波学习心得(1)方程推导
- 我的Qt学习之路——标准对话框
- Node实现简单的表单+图片上传+路由
- 随机抽取名字