aboutsummaryrefslogtreecommitdiff
path: root/lib
diff options
context:
space:
mode:
Diffstat (limited to 'lib')
-rw-r--r--lib/kmean.py46
-rw-r--r--lib/rerank.py83
2 files changed, 129 insertions, 0 deletions
diff --git a/lib/kmean.py b/lib/kmean.py
new file mode 100644
index 0000000..f041b19
--- /dev/null
+++ b/lib/kmean.py
@@ -0,0 +1,46 @@
+from logging import debug
+import numpy as np
+from sklearn.cluster import KMeans
+
+def create_kmean_clusters(feature_vectors, labels, gallery_idx, camId):
+ gallery = ([],[])
+ gallerylab = ([],[])
+ for i in range(gallery_idx.size):
+ cam = camId[gallery_idx[i]] - 1
+ gallery[cam].append(feature_vectors[gallery_idx[i]])
+ gallerylab[cam].append(labels[gallery_idx[i]])
+
+
+ train = np.array(gallery)
+ tlabel = np.array(gallerylab)
+
+ km_train_data = []
+ km_idx = []
+ km_labels = []
+
+ for i in range(2):
+ km_train_data.append(KMeans(n_clusters=int(np.max(labels)),random_state=0).fit(train[i]))
+ km_idx.append(km_train_data[i].labels_)
+ km_labels.append(range(np.max(labels)))
+ for j in range(np.max(labels)):
+ class_vote = np.zeros(np.max(labels))
+ for q in range(km_idx[i].size):
+ if km_idx[i][q]==j:
+ class_vote[int(tlabel[i][q])-1] += 1
+ km_labels[i][j] = np.argmax(class_vote) + 1
+
+ #MERGE CLUSTERS
+ cl = []
+ cllab = []
+ clustercam = []
+ for i in range(2):
+ clustercam.append(np.add(np.ones(len(km_labels[i])), i))
+ for j in range(len(km_labels[i])):
+ cl.append(km_train_data[i].cluster_centers_[j])
+ cllab.append(km_labels[i][j])
+
+ train_data = np.array(cl)
+ debug("Kmean data has shape", train_data.shape)
+ train_label = np.array(cllab)
+ train_cam = np.array([clustercam[i] for i in range(2)]).reshape(train_label.shape[0])
+ return train_data, train_label, train_cam
diff --git a/lib/rerank.py b/lib/rerank.py
new file mode 100644
index 0000000..6fb5b7b
--- /dev/null
+++ b/lib/rerank.py
@@ -0,0 +1,83 @@
+from scipy.spatial.distance import cdist
+import numpy as np
+from logging import debug
+
+def re_ranking(probFea,galFea,k1,k2,lambda_value, MemorySave = False, Minibatch = 2000):
+
+ query_num = probFea.shape[0]
+ all_num = query_num + galFea.shape[0]
+ feat = np.append(probFea,galFea,axis = 0)
+ feat = feat.astype(np.float16)
+ debug('computing original distance')
+ if MemorySave:
+ original_dist = np.zeros(shape = [all_num,all_num],dtype = np.float16)
+ i = 0
+ while True:
+ it = i + Minibatch
+ if it < np.shape(feat)[0]:
+ original_dist[i:it,] = np.power(cdist(feat[i:it,],feat),2).astype(np.float16)
+ else:
+ original_dist[i:,:] = np.power(cdist(feat[i:,],feat),2).astype(np.float16)
+ break
+ i = it
+ else:
+ original_dist = cdist(feat,feat).astype(np.float16)
+ original_dist = np.power(original_dist,2).astype(np.float16)
+ del feat
+ gallery_num = original_dist.shape[0]
+ original_dist = np.transpose(original_dist/np.max(original_dist,axis = 0))
+ V = np.zeros_like(original_dist).astype(np.float16)
+ initial_rank = np.argsort(original_dist).astype(np.int32)
+
+
+ debug('starting re_ranking')
+ for i in range(all_num):
+ # k-reciprocal neighbors
+ forward_k_neigh_index = initial_rank[i,:k1+1]
+ backward_k_neigh_index = initial_rank[forward_k_neigh_index,:k1+1]
+ fi = np.where(backward_k_neigh_index==i)[0]
+ k_reciprocal_index = forward_k_neigh_index[fi]
+ k_reciprocal_expansion_index = k_reciprocal_index
+ for j in range(len(k_reciprocal_index)):
+ candidate = k_reciprocal_index[j]
+ candidate_forward_k_neigh_index = initial_rank[candidate,:int(np.around(k1/2))+1]
+ candidate_backward_k_neigh_index = initial_rank[candidate_forward_k_neigh_index,:int(np.around(k1/2))+1]
+ fi_candidate = np.where(candidate_backward_k_neigh_index == candidate)[0]
+ candidate_k_reciprocal_index = candidate_forward_k_neigh_index[fi_candidate]
+ if len(np.intersect1d(candidate_k_reciprocal_index,k_reciprocal_index))> 2/3*len(candidate_k_reciprocal_index):
+ k_reciprocal_expansion_index = np.append(k_reciprocal_expansion_index,candidate_k_reciprocal_index)
+
+ k_reciprocal_expansion_index = np.unique(k_reciprocal_expansion_index)
+ weight = np.exp(-original_dist[i,k_reciprocal_expansion_index])
+ V[i,k_reciprocal_expansion_index] = weight/np.sum(weight)
+ original_dist = original_dist[:query_num,]
+ if k2 != 1:
+ V_qe = np.zeros_like(V,dtype=np.float16)
+ for i in range(all_num):
+ V_qe[i,:] = np.mean(V[initial_rank[i,:k2],:],axis=0)
+ V = V_qe
+ del V_qe
+ del initial_rank
+ invIndex = []
+ for i in range(gallery_num):
+ invIndex.append(np.where(V[:,i] != 0)[0])
+
+ jaccard_dist = np.zeros_like(original_dist,dtype = np.float16)
+
+ for i in range(query_num):
+ temp_min = np.zeros(shape=[1,gallery_num],dtype=np.float16)
+ indNonZero = np.where(V[i,:] != 0)[0]
+ indImages = []
+ indImages = [invIndex[ind] for ind in indNonZero]
+ for j in range(len(indNonZero)):
+ temp_min[0,indImages[j]] = temp_min[0,indImages[j]]+ np.minimum(V[i,indNonZero[j]],V[indImages[j],indNonZero[j]])
+ jaccard_dist[i] = 1-temp_min/(2-temp_min)
+
+ final_dist = jaccard_dist*(1-lambda_value) + original_dist*lambda_value
+ del original_dist
+ del V
+ del jaccard_dist
+ final_dist = final_dist[:query_num,query_num:]
+
+ return final_dist
+