DL杂记：YOLOV3之禅

from os import listdir
from os.path import isfile, join
import argparse
#import cv2
import numpy as np
import sys
import os
import shutil
import random 
import math
 
def IOU(x,centroids):
    '''
    :param x: 某一个ground truth的w,h
    :param centroids:  anchor的w,h的集合[(w,h),(),...]，共k个
    :return: 单个ground truth box与所有k个anchor box的IoU值集合
    '''
    IoUs = []
    w, h = x  # ground truth的w,h
    for centroid in centroids:
        c_w,c_h = centroid   #anchor的w,h
        if c_w>=w and c_h>=h:   #anchor包围ground truth
            iou = w*h/(c_w*c_h)
        elif c_w>=w and c_h<=h:    #anchor宽矮
            iou = w*c_h/(w*h + (c_w-w)*c_h)
        elif c_w<=w and c_h>=h:    #anchor瘦长
            iou = c_w*h/(w*h + c_w*(c_h-h))
        else: #ground truth包围anchor     means both w,h are bigger than c_w and c_h respectively
            iou = (c_w*c_h)/(w*h)
        IoUs.append(iou) # will become (k,) shape
    return np.array(IoUs)
 
def avg_IOU(X,centroids):
    '''
    :param X: ground truth的w,h的集合[(w,h),(),...]
    :param centroids: anchor的w,h的集合[(w,h),(),...]，共k个
    '''
    n,d = X.shape
    sum = 0.
    for i in range(X.shape[0]):
        sum+= max(IOU(X[i],centroids))  #返回一个ground truth与所有anchor的IoU中的最大值
    return sum/n    #对所有ground truth求平均
 
def write_anchors_to_file(centroids,X,anchor_file,input_shape,yolo_version):
    '''
    :param centroids: anchor的w,h的集合[(w,h),(),...]，共k个
    :param X: ground truth的w,h的集合[(w,h),(),...]
    :param anchor_file: anchor和平均IoU的输出路径
    '''
    f = open(anchor_file,'w')
    
    anchors = centroids.copy()
    print(anchors.shape)
 
    if yolo_version=='yolov2':
        for i in range(anchors.shape[0]):
            #yolo中对图片的缩放倍数为32倍，所以这里除以32，
            # 如果网络架构有改变，根据实际的缩放倍数来
            #求出anchor相对于缩放32倍以后的特征图的实际大小（yolov2）
            anchors[i][0]*=input_shape/32.
            anchors[i][1]*=input_shape/32.
    elif yolo_version=='yolov3':
        for i in range(anchors.shape[0]):
            #求出yolov3相对于原图的实际大小
            anchors[i][0]*=input_shape
            anchors[i][1]*=input_shape
    else:
        print("the yolo version is not right!")
        exit(-1)
 
    widths = anchors[:,0]
    sorted_indices = np.argsort(widths)
 
    print('Anchors = ', anchors[sorted_indices])
        
    for i in sorted_indices[:-1]:
        f.write('%0.2f,%0.2f, '%(anchors[i,0],anchors[i,1]))
 
    #there should not be comma after last anchor, that's why
    f.write('%0.2f,%0.2f\n'%(anchors[sorted_indices[-1:],0],anchors[sorted_indices[-1:],1]))
    
    f.write('%f\n'%(avg_IOU(X,centroids)))
    print()
 
def kmeans(X,centroids,eps,anchor_file,input_shape,yolo_version):
    
    N = X.shape[0] #ground truth的个数
    iterations = 0
    print("centroids.shape",centroids)
    k,dim = centroids.shape  #anchor的个数k以及w,h两维，dim默认等于2
    prev_assignments = np.ones(N)*(-1)    #对每个ground truth分配初始标签
    iter = 0
    old_D = np.zeros((N,k))  #初始化每个ground truth对每个anchor的IoU
 
    while True:
        D = []
        iter+=1           
        for i in range(N):
            d = 1 - IOU(X[i],centroids)
            D.append(d)
        D = np.array(D) # D.shape = (N,k)  得到每个ground truth对每个anchor的IoU
        
        print("iter {}: dists = {}".format(iter,np.sum(np.abs(old_D-D))))  #计算每次迭代和前一次IoU的变化值
            
        #assign samples to centroids 
        assignments = np.argmin(D,axis=1)  #将每个ground truth分配给距离d最小的anchor序号
        
        if (assignments == prev_assignments).all() :  #如果前一次分配的结果和这次的结果相同，就输出anchor以及平均IoU
            print("Centroids = ",centroids)
            write_anchors_to_file(centroids,X,anchor_file,input_shape,yolo_version)
            return
 
        #calculate new centroids
        centroid_sums=np.zeros((k,dim),np.float)   #初始化以便对每个簇的w,h求和
        for i in range(N):
            centroid_sums[assignments[i]]+=X[i]         #将每个簇中的ground truth的w和h分别累加
        for j in range(k):            #对簇中的w,h求平均
            centroids[j] = centroid_sums[j]/(np.sum(assignments==j)+1)
        
        prev_assignments = assignments.copy()     
        old_D = D.copy()  
 
def main(argv):
    parser = argparse.ArgumentParser()
    parser.add_argument('-filelist', default = r'train.txt',
                        help='path to filelist\n' )
    parser.add_argument('-output_dir', default = r'output/', type = str,
                        help='Output anchor directory\n' )
    parser.add_argument('-num_clusters', default = 0, type = int, 
                        help='number of clusters\n' )
    '''
    需要注意的是yolov2输出的值比较小是相对特征图来说的，
    yolov3输出值较大是相对原图来说的，
    所以yolov2和yolov3的输出是有区别的
    '''
    parser.add_argument('-yolo_version', default='yolov2', type=str,
                        help='yolov2 or yolov3\n')
    parser.add_argument('-yolo_input_shape', default=416, type=int,
                        help='input images shape，multiples of 32. etc. 416*416\n')
    args = parser.parse_args()
    
    if not os.path.exists(args.output_dir):
        os.mkdir(args.output_dir)
 
    f = open(args.filelist)
  
    lines = [line.rstrip('\n') for line in f.readlines()]
    
    annotation_dims = []
 
    for line in lines:
        line = line.replace('JPEGImages','labels')
        line = line.replace('.jpg','.txt')
        line = line.replace('.png','.txt')
        print(line)
        f2 = open(line)
        for line in f2.readlines():
            line = line.rstrip('\n')
            w,h = line.split(' ')[3:]            
            #print(w,h)
            annotation_dims.append((float(w),float(h)))
    annotation_dims = np.array(annotation_dims) #保存所有ground truth框的(w,h)
  
    eps = 0.005
 
    if args.num_clusters == 0:
        for num_clusters in range(1,11): #we make 1 through 10 clusters 
            anchor_file = join( args.output_dir,'anchors%d.txt'%(num_clusters))
 
            indices = [ random.randrange(annotation_dims.shape[0]) for i in range(num_clusters)]
            centroids = annotation_dims[indices]
            kmeans(annotation_dims,centroids,eps,anchor_file,args.yolo_input_shape,args.yolo_version)
            print('centroids.shape', centroids.shape)
    else:
        anchor_file = join( args.output_dir,'anchors%d.txt'%(args.num_clusters))
        indices = [ random.randrange(annotation_dims.shape[0]) for i in range(args.num_clusters)]
        centroids = annotation_dims[indices]
        kmeans(annotation_dims,centroids,eps,anchor_file,args.yolo_input_shape,args.yolo_version)
        print('centroids.shape', centroids.shape)
 
if __name__=="__main__":
    main(sys.argv)

def preprocess_true_boxes(true_boxes, true_labels, input_shape, anchors, num_classes):
    """
    Preprocess true boxes to training input format
    Parameters:
    -----------
    :param true_boxes: numpy.ndarray of shape [T, 4]
                        T: the number of boxes in each image.
                        4: coordinate => x_min, y_min, x_max, y_max
    :param true_labels: class id
    :param input_shape: the shape of input image to the yolov3 network, [416, 416]
    :param anchors: array, shape=[9,2], 9: the number of anchors, 2: width, height
    :param num_classes: integer, for coco dataset, it is 80
    Returns:
    ----------
    y_true: list(3 array), shape like yolo_outputs, [13, 13, 3, 85]
                           13:cell szie, 3:number of anchors
                           85: box_centers, box_sizes, confidence, probability
    """
    input_shape = np.array(input_shape, dtype=np.int32)
    num_layers = len(anchors) // 3
    anchor_mask = [[6,7,8], [3,4,5], [0,1,2]] if num_layers==3 else [[3,4,5], [1,2,3]]
    grid_sizes = [input_shape//32, input_shape//16, input_shape//8]

    box_centers = (true_boxes[:, 0:2] + true_boxes[:, 2:4]) / 2 # the center of box
    box_sizes =  true_boxes[:, 2:4] - true_boxes[:, 0:2] # the height and width of box

    true_boxes[:, 0:2] = box_centers
    true_boxes[:, 2:4] = box_sizes

    y_true_13 = np.zeros(shape=[grid_sizes[0][0], grid_sizes[0][1], 3, 5+num_classes], dtype=np.float32)
    y_true_26 = np.zeros(shape=[grid_sizes[1][0], grid_sizes[1][1], 3, 5+num_classes], dtype=np.float32)
    y_true_52 = np.zeros(shape=[grid_sizes[2][0], grid_sizes[2][1], 3, 5+num_classes], dtype=np.float32)

    y_true = [y_true_13, y_true_26, y_true_52]
    anchors_max =  anchors / 2.
    anchors_min = -anchors_max
    valid_mask = box_sizes[:, 0] > 0

    # Discard zero rows.
    wh = box_sizes[valid_mask]
    # set the center of all boxes as the origin of their coordinates
    # and correct their coordinates
    wh = np.expand_dims(wh, -2)
    boxes_max = wh / 2.
    boxes_min = -boxes_max

    intersect_mins = np.maximum(boxes_min, anchors_min)
    intersect_maxs = np.minimum(boxes_max, anchors_max)
    intersect_wh = np.maximum(intersect_maxs - intersect_mins, 0.)
    intersect_area = intersect_wh[..., 0] * intersect_wh[..., 1]
    box_area = wh[..., 0] * wh[..., 1]

    anchor_area = anchors[:, 0] * anchors[:, 1]
    iou = intersect_area / (box_area + anchor_area - intersect_area)
    # Find best anchor for each true box
    best_anchor = np.argmax(iou, axis=-1)

    for t, n in enumerate(best_anchor):
        for l in range(num_layers):
            if n not in anchor_mask[l]: continue
            # 这里也是用到归一化的思想。
            i = np.floor(true_boxes[t,1]/input_shape[::-1]*grid_sizes[l][0]).astype('int32')
            j = np.floor(true_boxes[t,0]/input_shape[::-1]*grid_sizes[l][1]).astype('int32')
            k = anchor_mask[l].index(n)
            c = true_labels[t].astype('int32')
            y_true[l][i, j, k, 0:4] = true_boxes[t, 0:4]
            y_true[l][i, j, k,   4] = 1
            y_true[l][i, j, k, 5+c] = 1

    return y_true_13, y_true_26, y_true_52

def preprocess_true_boxes(true_boxes, anchors, image_size):
 """
参数
--------------
true_boxes : 实际框的位置和类别，我们的输入。二个维度：
第一个维度：一张图片中有几个实际框
第二个维度： [x, y, w, h, class]，x,y 是框中心点坐标，w,h 是框的宽度和高度。x,y,w,h 均是除以图片
           分辨率得到的[0,1]范围的比值。
  
anchors : 实际anchor boxes 的值，论文中使用了五个。[w,h]，都是相对于gird cell 的比值。二个维度：
第一个维度：anchor boxes的数量，这里是5
第二个维度：[w,h]，w,h,都是相对于gird cell长宽的比值。
           [1.08, 1.19], [3.42, 4.41], [6.63, 11.38], [9.42, 5.11], [16.62, 10.52]
        
image_size : 图片的实际尺寸。这里是416x416。
Returns
--------------
detectors_mask : 取值是0或者1，这里的shape是[13,13,5,1]，四个维度。
第一个维度：true_boxes的中心位于第几行（y方向上属于第几个gird cell）
第二个维度：true_boxes的中心位于第几列（x方向上属于第几个gird cell）
第三个维度：哪个anchor box
第四个维度：0/1。1的就是用于预测改true boxes 的 anchor boxes

matching_true_boxes: 这里的shape是[13,13,5,5]，四个维度。
第一个维度：true_boxes的中心位于第几行（y方向上属于第几个gird cel）
第二个维度：true_boxes的中心位于第几列（x方向上属于第几个gird cel）
第三个维度：第几个anchor box
第四个维度：[x,y,w,h,class]。这里的x，y表示offset，是相当于gird cell的，w,h是取了log函数的，
class是属于第几类。后面的代码会详细看到
"""

    height, width = image_size
    num_anchors = len(anchors)

    assert height % 32 == 0,   '输入的图片的高度必须是32的倍数，不然会报错。'
    assert width % 32 == 0,   '输入的图片的宽度必须是32的倍数，不然会报错。'

    conv_height = height // 32    '进行gird cell划分'
    conv_width = width // 32    '进行gird cell划分'

    num_box_params = true_boxes.shape[1] 
    detectors_mask = np.zeros(
        (conv_height, conv_width, num_anchors, 1), dtype=np.float32)
    matching_true_boxes = np.zeros(
        (conv_height, conv_width, num_anchors, num_box_params),
        dtype=np.float32)    '确定detectors_mask和matching_true_boxes的维度，用0填充'

    for box in true_boxes:    '遍历实际框'
        box_class = box[4:5]    '提取类别信息，属于哪类'

        box = box[0:4] * np.array(
            [conv_width, conv_height, conv_width, conv_height])   '换算成相对于gird cell的值'

        i = np.floor(box[1]).astype('int')    '（y方向上属于第几个gird cell）'
        j = np.floor(box[0]).astype('int')    '（x方向上属于第几个gird cell）'
        best_iou = 0
        best_anchor = 0


        '计算anchor boxes 和 true boxes的iou，找到最佳预测的一个anchor boxes'
        for k, anchor in enumerate(anchors):
            # Find IOU between box shifted to origin and anchor box.
            box_maxes = box[2:4] / 2.
            box_mins = -box_maxes
            anchor_maxes = (anchor / 2.)
            anchor_mins = -anchor_maxes

            intersect_mins = np.maximum(box_mins, anchor_mins)
            intersect_maxes = np.minimum(box_maxes, anchor_maxes)
            intersect_wh = np.maximum(intersect_maxes - intersect_mins, 0.)
            intersect_area = intersect_wh[0] * intersect_wh[1]
            box_area = box[2] * box[3]
            anchor_area = anchor[0] * anchor[1]
            iou = intersect_area / (box_area + anchor_area - intersect_area)
            if iou > best_iou:
                best_iou = iou
                best_anchor = k


        if best_iou > 0:
            detectors_mask[i, j, best_anchor] = 1  '找到最佳预测anchor boxes'
            adjusted_box = np.array(
                [
                    box[0] - j, box[1] - i, 'x,y都是相对于gird cell的位置，左上角[0,0]，右下角[1,1]'
                    np.log(box[2] / anchors[best_anchor][0]),  '对应实际框w,h和anchor boxes w,h的比值取log函数'
                    np.log(box[3] / anchors[best_anchor][1]), box_class  'class实际框的物体是属于第几类'
                ],
                dtype=np.float32)
            matching_true_boxes[i, j, best_anchor] = adjusted_box   
    return detectors_mask, matching_true_boxes