---

title: Triplet Selection
grammar_cjkRuby: true
---

样本选取的理想情况：

a minimal number of exemplars of any one identity is present in each mini-batch

单个个体的全部样本必须存在于每个mini-batch中

In our experiments we sample the training data such that around 40 faces are selected per identity per minibatch. Additionally, randomly sampled negative faces are added to each mini-batch.

每个mini-batch中每个个体有40张图片，先组成A-P对，然后再随机抽取负样本到mini-batch中组成（A,P,N)三元组。

Instead of picking the hardest positive, we use all anchor-positive pairs in a mini-batch while still selecting the hard negatives. We found in practice that the all anchor-positive method was more stable and converged slightly faster（收敛更快） at the beginning of training

所有的A-P对和满足条件的A-N对

具体过程：
先从训练数据集中抽样出一组图片到mini-batch中，计算这些图片在当时的网络模型中得到的embedding，保存在emb_array当中。
每个mini-batch有多类图，每类图有多张

st=>start: mini-batch中第i类图的第j张图片
e=>end: 得到三元组（A.P.N)
op=>operation: 此图片和其他所有图片（这个mini-batch中的）的欧氏距离
op1=>operation: 此图片与同类其他图片组成(anchor,postive)图片，记为(A,P)
op2=>operation: 此图片与其他类的图片组成(anchor,negtive)图片，记为(A,N)st->op->op1->op2->e

（A,N)需满足的条件：

 all_neg = np.where(neg_dists_sqr-pos_dist_sqr<alpha)[0] # VGG Face

#多少人，alpha参数
def select_triplets(embeddings, nrof_images_per_class, image_paths, people_per_batch, alpha):""" Select the triplets for training"""trip_idx = 0#某个人的图片的embedding在emb_arr中的开始的索引emb_start_idx = 0num_trips = 0triplets = []# VGG Face: Choosing good triplets is crucial and should strike a balance between#  selecting informative (i.e. challenging) examples and swamping training with examples that#  are too hard. This is achieve by extending each pair (a, p) to a triplet (a, p, n) by sampling#  the image n at random, but only between the ones that violate the triplet loss margin. The#  latter is a form of hard-negative mining, but it is not as aggressive (and much cheaper) than#  choosing the maximally violating example, as often done in structured output learning.#遍历每一个人for i in xrange(people_per_batch):#这个人有多少张图片nrof_images = int(nrof_images_per_class[i])#遍历第i个人的所有图片for j in xrange(1,nrof_images):#第j张图的embedding在emb_arr 中的位置a_idx = emb_start_idx + j - 1#第j张图跟其他所有图片的欧氏距离neg_dists_sqr = np.sum(np.square(embeddings[a_idx] - embeddings), 1)#遍历每一对可能的(anchor,postive)图片，记为(a,p)吧for pair in xrange(j, nrof_images): # For every possible positive pair.#第p张图片在emb_arr中的位置p_idx = emb_start_idx + pair#(a,p)之前的欧式距离pos_dist_sqr = np.sum(np.square(embeddings[a_idx]-embeddings[p_idx]))#同一个人的图片不作为negative，所以将距离设为无穷大neg_dists_sqr[emb_start_idx:emb_start_idx+nrof_images] = np.NaN#all_neg = np.where(np.logical_and(neg_dists_sqr-pos_dist_sqr<alpha, pos_dist_sqr<neg_dists_sqr))[0]  # FaceNet selection#其他人的图片中有哪些图片与a之间的距离-p与a之间的距离小于alpha的all_neg = np.where(neg_dists_sqr-pos_dist_sqr<alpha)[0] # VGG Face selecction#所有可能的negativenrof_random_negs = all_neg.shape[0]#如果有满足条件的negativeif nrof_random_negs>0:#从中随机选取一个作为nrnd_idx = np.random.randint(nrof_random_negs)n_idx = all_neg[rnd_idx]# 选到(a,p,n)作为三元组triplets.append((image_paths[a_idx], image_paths[p_idx], image_paths[n_idx]))#print('Triplet %d: (%d, %d, %d), pos_dist=%2.6f, neg_dist=%2.6f (%d, %d, %d, %d, %d)' % #    (trip_idx, a_idx, p_idx, n_idx, pos_dist_sqr, neg_dists_sqr[n_idx], nrof_random_negs, rnd_idx, i, j, emb_start_idx))trip_idx += 1num_trips += 1emb_start_idx += nrof_imagesnp.random.shuffle(triplets)return triplets, num_trips, len(triplets)

更多代码解释请看：http://www.cnblogs.com/irenelin/p/7831338.html