test code

main.py

@@ -2,11 +2,20 @@ from torch.utils.data import DataLoader
 from data.CUB_200 import get_dataset, input_transform, input_transform2
 import SCDA
 from torchvision import models 
+import pandas as pd
 import torch
 import torch.nn.functional as F
 import csv
 import numpy as np
 from util.model import pool_model
+import random
+
+
+def retrieve(q, data, num=5):
+    distances = np.sum(np.square((data - q)), axis=-1)
+    # distances = np.sum(cosine_similarity(q, data)), axis=-1)
+    indices = distances.argsort(axis=0)[:num]
+    return indices
 
 
 def write_csv(results,file_name):
@@ -26,54 +35,116 @@ data_or = DataLoader(dataset_or, batch_size=10, shuffle=True,num_workers=0)
 net1.eval()
 net2.eval()
 result = []
+label_re = []
 max_ave_pool = pool_model()
-out = open('feat.csv', 'a', newline='')
+# out = open('feat.csv', 'a', newline='')
-csv_write = csv.writer(out, dialect='excel')
+# csv_write = csv.writer(out, dialect='excel')
-csv_write.writerow(['features', 'label'])
+# csv_write.writerow(['features', 'label'])
 for ii, (img1, img2, label) in enumerate(data_or):
+    # if ii == 2:
+        # exit()
     input1 = img1
     feat_re = net1(input1)
     feat_po = net2(input1)
     input2 = img2
     feat_flip_re = net1(input2)
     feat_flip_po = net2(input2)
-    m, _, _, _ = feat_re.shape
+    m, _, h1, w1 = feat_po.shape
+    m, _, h2, w2 = feat_re.shape
     label = label.detach().numpy()
     f_po = torch.zeros(feat_po.shape)
     f_re = torch.zeros(feat_re.shape)
     filp_po = torch.zeros(feat_flip_po.shape)
     filp_re = torch.zeros(feat_flip_re.shape)
+    cc8 = np.zeros((m, h1, w1)) #31
+    cc7 = np.zeros((m, h1, w1)) # 31
+    cc6 = np.zeros((m, h2, w2))# 28
+    cc5 = np.zeros((m, h2, w2))# 28
+    
+
     for i in range(m):
         # f_re[i] = SCDA.select_aggregate(feat_re[i].detach().numpy())
         f_po[i] = SCDA.select_aggregate(feat_po[i])[0] # 31a
         cc2 = SCDA.select_aggregate(feat_po[i])[1]
-        f_re[i] = SCDA.select_aggregate_and(feat_re[i], cc2) # 28a
+        cc8[i] = cc2
+        f_re[i] = SCDA.select_aggregate_and(feat_re[i], cc2)[0] # 28a
+        cc3 = SCDA.select_aggregate_and(feat_re[i], cc2)[1]
+        cc6[i] = cc3
 
         
         filp_po[i] = SCDA.select_aggregate(feat_flip_po[i])[0] # 31b
         cc2_f = SCDA.select_aggregate(feat_flip_po[i])[1]
-        filp_re[i] = SCDA.select_aggregate_and(feat_flip_re[i], cc2_f) # 28b
+        cc7[i] = cc2_f
+        filp_re[i] = SCDA.select_aggregate_and(feat_flip_re[i], cc2_f)[0]
+        cc3_f = SCDA.select_aggregate_and(feat_flip_re[i], cc2_f)[1] # 28b
+        cc5[i] = cc3_f
         # print(f.shape)
         # l = int(lable[i])
         # csv_write.writerow([f, l])
         # del f, l
         # result.append((f, l))
     # print(f_po.shape)
-    l31a = max_ave_pool(f_po)
+    l31a = max_ave_pool(f_po, cc8)
-    l28a = max_ave_pool(f_re)
+    # exit()
+    l28a = max_ave_pool(f_re, cc6)
 
-    l31b = max_ave_pool(filp_po)
+    l31b = max_ave_pool(filp_po, cc7)
-    l28b = max_ave_pool(filp_re)
+    l28b = max_ave_pool(filp_re, cc5)
+    # print(f_po.reshape(1, -1))
+    # print(f_re)
+    # print(filp_po)
+    # print(filp_re)
 
     feat = torch.cat((l31a, 0.5*l28a, l31b, 0.5*l28b), dim=1).reshape(m, -1).detach().numpy()
+    # print(feat.shape)
 
     for i in range(m):
-        csv_write.writerow([feat[i], int(label[i])])
+        # print(feat[i])
+        # print(feat[i].reshape(1,-1).shape)
+        result.append(feat[i].tolist())
+        label_re.append(int(label[i]))
+        # exit()
         
-    print('batch {}/{} complete'.format(ii, len(data_or)))
+        # az = feat[i].tolist()
+        # print(az)
+        # csv_write.writerow([az, int(label[i])])
  
+    print('batch {}/{} complete'.format(ii+1, len(data_or)))
 
+print("test...")
 
+feats = np.vstack(result)
+# print(feats.shape)
+label = np.vstack(label_re)
+
+x = [int(i) for i in range(feats.shape[0])]
+random.shuffle(x)
+feats = feats[x]
+labels = label[x]
+# labels = [i[0] for i in label]
+test_data = feats[:int(0.5*len(labels))]
+train_data = feats[int(0.5*len(labels)):]
+test_label = labels[:int(0.5*len(labels))]
+train_label = labels[int(0.5*len(labels)):]
+
+top1 = 0
+top5 = 0
+# ap = 0.0
+for i in range(len(test_label)):
+    # print(test_data[i].shape)
+    # exit()
+    inds = retrieve(test_data[i].reshape(1,-1), train_data, num=len(train_data))
+    # print(inds.tolist())
+    # print(inds)
+    labels = train_label[inds]
+    if labels[0] == test_label[i]:
+        top1 +=1
+    if test_label[i] in labels[:5]:
+        top5 +=1
+    # ap += compute_ap(inds.tolist(), get_gt(test_label[i], train_labels))
+    print("Query[%d / %d] complete: top1: %.4f, top5: %.4f"%(i, len(test_label), top1/(i+1), top5/(i+1)))
+
+print("top1 acc: %.4f, top5 acc %.4f"%(top1/len(test_label), top5/len(test_label)))
 
 # avg.train_data_L31a
 # maxi.train_data_L31a