跨模态检索的mAP（@R）和 PR曲线（Precision-Re

对于基础知识网上资料很多，对此不在重述。本文主要是记录pytorch下怎么实现。

1.mAP（@R）（参考DCMH：https://github.com/WendellGul/DCMH/blob/master/utils.py)

import torch

def calc_hammingDist(B1, B2):
  q = B2.shape[1]
  if len(B1.shape) < 2:
      B1 = B1.unsqueeze(0)
  distH = 0.5 * (q - B1.mm(B2.transpose(0, 1)))
  return distH


def calc_map_k(qB, rB, query_L, retrieval_L, k=None):
  # qB: {-1,+1}^{mxq}
  # rB: {-1,+1}^{nxq}
  # query_L: {0,1}^{mxl}
  # retrieval_L: {0,1}^{nxl}
  num_query = query_L.shape[0]
  map = 0
  if k is None:
      k = retrieval_L.shape[0]
  for iter in range(num_query):
      q_L = query_L[iter]
      if len(q_L.shape) < 2:
          q_L = q_L.unsqueeze(0)
      gnd = (q_L.mm(retrieval_L.transpose(0, 1)) > 0).squeeze().type(torch.float32)
      tsum = torch.sum(gnd)
      if tsum == 0:
          continue
      hamm = calc_hammingDist(qB[iter, :], rB)
      _, ind = torch.sort(hamm)
      ind.squeeze_()
      gnd = gnd[ind]
      total = min(k, int(tsum))
      count = torch.arange(1, total + 1).type(torch.float32)
      tindex = torch.nonzero(gnd)[:total].squeeze().type(torch.float32) + 1.0
      if tindex.is_cuda:
          count = count.cuda()
      map = map + torch.mean(count / tindex)
  map = map / num_query
  return map

2.Precision-Recall Curve（参考：https://blog.csdn.net/HackerTom/article/details/89425729）

import matplotlib.pyplot as plt
def pr_curve(qB, rB, qL, rL, ep, task,  topK=-1):
  n_query = qB.shape[0]
  if topK == -1 or topK > rB.shape[0]:  # top-K 之 K 的上限
    topK = rB.shape[0]


  # Gnd = (np.dot(qL, rL.transpose()) > 0).astype(np.float32)
  Gnd = (qL.mm(rL.transpose(0, 1)) > 0).type(torch.float32)
  _,Rank =  torch.sort(calc_hammingDist(qB, rB))
  P, R = [], []
  # KK = []
  # K_ = [x * 2000 + 1 for x in range(1, int(topK/2000))]
  # for i in K_:
  #     if i < topK:
  #         KK.append(i)
  for k in range(1, topK+1):  # 枚举 top-K 之 K
      # ground-truth: 1 vs all
      p = torch.zeros(n_query)  # 各 query sample 的 Precision@R
      r = torch.zeros(n_query)  # 各 query sample 的 Recall@R
      for it in range(n_query):  # 枚举 query sample

          gnd = Gnd[it]
          gnd_all = torch.sum(gnd)  # 整个被检索数据库中的相关样本数
          if gnd_all == 0:
              continue

          asc_id = Rank[it][:k]
          gnd = gnd[asc_id]
          gnd_r = torch.sum(gnd)  # top-K 中的相关样本数
          p[it] = gnd_r / k
          r[it] = gnd_r / gnd_all

      P.append(torch.mean(p))
      R.append(torch.mean(r))
  print(P)
  print(R)

# 画 P-R 曲线
fig = plt.figure(figsize=(5, 5))
plt.plot(R, P)  # 第一个是 x，第二个是 y
plt.grid(True)
plt.xlim(0, 1)
plt.ylim(0, 1)
plt.xlabel('recall')
plt.ylabel('precision')
plt.legend()
plt.show()

3.实列说明

if __name__ == '__main__':
  qB = torch.Tensor([[1, -1, 1, 1],
                     [-1, -1, -1, 1],
                     [1, 1, -1, 1],
                     [1, 1, 1, -1]])
  rB = torch.Tensor([[1, -1, 1, -1],
                     [-1, -1, 1, -1],
                     [-1, -1, 1, -1],
                     [1, 1, -1, -1],
                     [-1, 1, -1, -1],
                     [1, 1, -1, 1]])
  query_L = torch.Tensor([[0, 1, 0, 0],
                          [1, 1, 0, 0],
                          [1, 0, 0, 1],
                          [0, 1, 0, 1]])
  retrieval_L = torch.Tensor([[1, 0, 0, 1],
                              [1, 1, 0, 0],
                              [0, 1, 1, 0],
                              [0, 0, 1, 0],
                              [1, 0, 0, 0],
                              [0, 0, 1, 0]])

  map = calc_map_k(qB, rB, query_L, retrieval_L)
  print("map", map)
  pr = pr_curve(qB, rB, query_L, retrieval_L, 2, 'i2t',  topK=-1)
  print('pr', pr)

4.结果

map tensor(0.7042)
[tensor(0.5000), tensor(0.5000), tensor(0.6667), tensor(0.6250), tensor(0.6000), tensor(0.5000)]
[tensor(0.1458), tensor(0.3333), tensor(0.6875), tensor(0.8542), tensor(1.), tensor(1.)]