From 98494a3055b68be4bd60d409052a0c0117dfca3e Mon Sep 17 00:00:00 2001 From: Vasil Zlatanov Date: Sun, 2 Dec 2018 18:14:01 +0000 Subject: Rewrite code to exclude images from same cam and label --- part2.py | 282 +++++++++++++++++--------------------------------------------- rerank.py | 82 ++++++++++++++++++ 2 files changed, 158 insertions(+), 206 deletions(-) create mode 100644 rerank.py diff --git a/part2.py b/part2.py index 299fdcd..1792f81 100755 --- a/part2.py +++ b/part2.py @@ -1,11 +1,9 @@ -#!/usr/bin/env python +#!/usr/bin/python -W ignore::DeprecationWarning # Author: Vasil Zlatanov, Nunzio Pucci # EE4 Pattern Recognition coursework # -# usage: train.py [-h] -i DATA [-m EIGEN] [-M REIGEN] [-e ENSEMBLE] [-b] -# [-R RANDOM] [-n NEIGHBORS] [-f FACES] [-c] [-s SEED] -# [-t SPLIT] [-2] [-p] [-l] [-r RECONSTRUCT] [-cm] [-q] [-pr] -# [-alt] +# usage: part2.py [-h] [-t] [-cm] [-km] [-ma] [-e] [-r] [-ka RERANKA] +# [-kb RERANKB] [-v] import matplotlib.pyplot as plt from mpl_toolkits.mplot3d import Axes3D @@ -30,90 +28,28 @@ from numpy import genfromtxt from numpy import linalg as LA from timeit import default_timer as timer from scipy.spatial.distance import cdist +from rerank import re_ranking -#prob query, gal train -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) - print('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) - - - print('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) +parser = argparse.ArgumentParser() +parser.add_argument("-t", "--test", help="Use test data instead of query", action='store_true') +parser.add_argument("-cm", "--conf_mat", help="Show visual confusion matrix", action='store_true') +parser.add_argument("-km", "--kmean", help="Perform Kmeans", action='store_true', default=0) +parser.add_argument("-ma", "--mahalanobis", help="Perform Mahalanobis Distance metric", action='store_true', default=0) +parser.add_argument("-e", "--euclidean", help="Standard euclidean", action='store_true', default=0) +parser.add_argument("-r", "--rerank", help="Use k-reciprocal rernaking", action='store_true') +parser.add_argument("-ka", "--reranka", help="Parameter 1 for Rerank", type=int, default = 20) +parser.add_argument("-kb", "--rerankb", help="Parameter 2 for rerank", type=int, default = 6) +parser.add_argument("-v", "--verbose", help="Use verbose output", action='store_true') +args = parser.parse_args() - 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 +def verbose(*text): + if args.verbose: + print(text) -def draw_results(args, target_test, target_pred): - acc_sc = accuracy_score(target_test, target_pred) - cm = confusion_matrix(target_test, target_pred) +#prob query, gal train +def draw_results(test_label, pred_label): + acc_sc = accuracy_score(test_label, pred_label) + cm = confusion_matrix(test_label, pred_label) print('Accuracy: ', acc_sc) if (args.conf_mat): plt.matshow(cm, cmap='Blues') @@ -123,121 +59,62 @@ def draw_results(args, target_test, target_pred): plt.show() return -def test_model(train_data, test_data, target_train, target_test, args): - classifier = KNeighborsClassifier(n_neighbors=args.neighbors, metric='euclidean') -# else: -# S = LA.inv(np.cov(train_data, rowvar=False)) -# print(S.shape) -# classifier = KNeighborsClassifier(n_neighbors=args.neighbors, metric='mahalanobis', metric_params={'VI':S}) - classifier.fit(train_data, target_train) - target_pred = classifier.predict(test_data) - dist, nn_idx = classifier.kneighbors(test_data) - #USE NN_IDX TO RECOVER NEIGHBORS +def test_model(gallery_data, probe_data, gallery_label, probe_label, gallery_cam, probe_cam): + # metric = 'jaccard' is also valid + if args.mahalanobis: + metric = 'sqeuclidean' + else: + metric = 'euclidean' + + verbose("probe shape:", probe_data.shape) + verbose("gallery shape:", gallery_data.shape) + + if args.rerank: + distances = re_ranking(probe_data, gallery_data, + args.reranka ,args.rerankb , 0.3, + MemorySave = False, Minibatch = 2000) + else: + distances = cdist(probe_data, gallery_data, metric) + + ranklist = np.argsort(distances, axis=1) + + target_pred = np.zeros(ranklist.shape[0]) + for probe_idx in range(probe_data.shape[0]): + row = ranklist[probe_idx] + n = 0 + while (probe_cam[probe_idx] == gallery_cam[row[n]] and + probe_label[probe_idx] == gallery_label[row[n]]): + n += 1 + target_pred[probe_idx] = gallery_label[row[n]] + return target_pred def main(): - parser = argparse.ArgumentParser() - parser.add_argument("-R", "--random", help="Number of eigen value to randomise", type=int) - parser.add_argument("-n", "--neighbors", help="How many neighbors to use", type=int, default = 1) - parser.add_argument("-c", "--principal", help="Show principal components", action='store_true') - parser.add_argument("-s", "--seed", help="Seed to use", type=int, default=0) - parser.add_argument("-t", "--split", help="Fractoin of data to use for testing", type=float, default=0.3) - parser.add_argument("-2", "--grapheigen", help="Swow 2D graph of targets versus principal components", - action='store_true') - parser.add_argument("-cm", "--conf_mat", help="Show visual confusion matrix", action='store_true') - parser.add_argument("-q", "--pca_r", help="Use Reduced PCA", action='store_true') - parser.add_argument("-pr", "--prob", help="Certainty on each guess", action='store_true') - parser.add_argument("-km", "--kmean", help="Perform Kmeans", action='store_true', default=0) - parser.add_argument("-ma", "--mala", help="Perform Mahalanobis Distance metric", action='store_true', default=0) - parser.add_argument("-e", "--eucl", help="Standard euclidean", action='store_true', default=0) - parser.add_argument("-ka", "--reranka", help="Parameter 1 for Rerank", type=int, default = 20) - parser.add_argument("-kb", "--rerankb", help="Parameter 2 for rerank", type=int, default = 6) - args = parser.parse_args() - - ###PART2 INPUT DATA mat = scipy.io.loadmat('data/cuhk03_new_protocol_config_labeled.mat') camId = mat['camId'] filelist = mat['filelist'] - gallery_idx = mat['gallery_idx'] labels = mat['labels'] - query_idx = mat['query_idx'] - train_idx = mat['train_idx'] + gallery_idx = mat['gallery_idx'] - 1 + query_idx = mat['query_idx'] - 1 + train_idx = mat['train_idx'] - 1 with open("data/feature_data.json", "r") as read_file: feature_vectors = np.array(json.load(read_file)) + + gallery_idx = gallery_idx.reshape(gallery_idx.shape[0]) + if args.test: + query_idx = train_idx.reshape(train_idx.shape[0]) + else: + query_idx = query_idx.reshape(query_idx.shape[0]) + camId = camId.reshape(camId.shape[0]) + + train_data = feature_vectors[gallery_idx] + test_data = feature_vectors[query_idx] + train_label = labels[gallery_idx] + test_label = labels[query_idx] + train_cam = camId[gallery_idx] + test_cam = camId[query_idx] - query_cam_1 = 0 - for i in range(query_idx.size): - if camId[query_idx[i]] == 1: - query_cam_1 = query_cam_1 + 1 - query_cam_2 = query_idx.size - query_cam_1 - - train_cam_1 = 0 - for i in range(gallery_idx.size): - if camId[gallery_idx[i]] == 1: - train_cam_1 = train_cam_1 + 1 - train_cam_2 = gallery_idx.size - train_cam_1 - - train_data_1 = np.zeros(((train_cam_1),(feature_vectors.shape[1]))) - train_label_1 = np.zeros(train_cam_1) - test_data_1 = np.zeros(((query_cam_1),(feature_vectors.shape[1]))) - test_label_1 = np.zeros(query_cam_1) - - train_data_2 = np.zeros(((train_cam_2),(feature_vectors.shape[1]))) - train_label_2 = np.zeros(train_cam_2) - test_data_2 = np.zeros(((query_cam_2),(feature_vectors.shape[1]))) - test_label_2 = np.zeros(query_cam_2) - - i_1 = 0 - i_2 = 0 - for i in range(gallery_idx.size): - if camId[gallery_idx[i]] == 1: - train_data_1[i_1] = feature_vectors[gallery_idx[i]] - i_1 = i_1 + 1 - else: - train_data_2[i_2] = feature_vectors[gallery_idx[i]] - i_2 = i_2 + 1 - i_1 = 0 - i_2 = 0 - for i in range(query_idx.size): - if camId[query_idx[i]] == 1: - test_data_1[i_1] = feature_vectors[query_idx[i]] - i_1 = i_1 + 1 - else: - test_data_2[i_2] = feature_vectors[query_idx[i]] - i_2 = i_2 + 1 - i_1 = 0 - i_2 = 0 - for i in range(gallery_idx.size): - if camId[gallery_idx[i]] == 1: - train_label_1[i_1] = labels[gallery_idx[i]] - i_1 = i_1 + 1 - else: - train_label_2[i_2] = labels[gallery_idx[i]] - i_2 = i_2 + 1 - i_1 = 0 - i_2 = 0 - for i in range(query_idx.size): - if camId[query_idx[i]] == 1: - test_label_1[i_1] = labels[query_idx[i]] - i_1 = i_1 + 1 - else: - test_label_2[i_2] = labels[query_idx[i]] - i_2 = i_2 + 1 - - if (args.mala): - final_dist = re_ranking(test_data_1, train_data_2, args.reranka, args.rerankb, 0.3) - target_pred = np.zeros(final_dist.shape[0]) - for i in range(test_label_1.size): - target_pred[i] = train_label_2[np.argmin(final_dist[i])] - draw_results(args, test_label_1, target_pred) - - final_dist2 = re_ranking(test_data_2, train_data_1, args.reranka, args.rerankb, 0.3) - target_pred2 = np.zeros(final_dist2.shape[0]) - for i in range(test_label_2.size): - target_pred2[i] = train_label_1[np.argmin(final_dist2[i])] - draw_results(args, test_label_2, target_pred2) - - elif(args.kmean): + if(args.kmean): km_labels_1 = np.arange(1,np.max(labels)+1) km_labels_2 = np.arange(1,np.max(labels)+1) km_train_data_1 = np.zeros(((km_labels_1.size),(feature_vectors.shape[1]))) @@ -253,8 +130,8 @@ def main(): class_vote[int(train_label_1[q])-1] = class_vote[int(train_label_1[q])-1] + 1 km_labels_1[i] = np.argmax(class_vote) + 1 - target_pred = test_model(km_train_data_1.cluster_centers_, test_data_2, km_labels_1, test_label_2, args) - draw_results(args, test_label_2, target_pred) + target_pred = test_model(km_train_data_1.cluster_centers_, test_data_2, km_labels_1, test_label_2) + draw_results(test_label_2, target_pred) km_idx_2 = km_train_data_2.labels_ for i in range(np.max(labels)): @@ -264,20 +141,13 @@ def main(): class_vote[int(train_label_2[q])-1] = class_vote[int(train_label_2[q])-1] + 1 km_labels_2[i] = np.argmax(class_vote) + 1 - target_pred = test_model(km_train_data_2.cluster_centers_, test_data_1, km_labels_2, test_label_1, args) - draw_results(args, test_label_1, target_pred) + target_pred = test_model(km_train_data_2.cluster_centers_, test_data_1, km_labels_2, test_label_1) + draw_results(test_label_1, target_pred) - elif(args.eucl): - target_pred = test_model(train_data_2, test_data_1, train_label_2, test_label_1, args) - draw_results(args, test_label_1, target_pred) - target_pred = test_model(train_data_1, test_data_2, train_label_1, test_label_2, args) - draw_results(args, test_label_2, target_pred) - - - print('N-Query from cam 1:', test_data_1.shape) - print('N-Query from cam 2:', test_data_2.shape) - print('Complete') + else: + target_pred = test_model(train_data, test_data, train_label, test_label, train_cam, test_cam) + draw_results(test_label, target_pred) if __name__ == "__main__": main() - \ No newline at end of file + diff --git a/rerank.py b/rerank.py new file mode 100644 index 0000000..6b20f53 --- /dev/null +++ b/rerank.py @@ -0,0 +1,82 @@ +from scipy.spatial.distance import cdist +import numpy as np + +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) + print('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) + + + print('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 + -- cgit v1.2.3-54-g00ecf