|
- # Copyright 2021 Huawei Technologies Co., Ltd
- #
- # Licensed under the Apache License, Version 2.0 (the "License");
- # you may not use this file except in compliance with the License.
- # You may obtain a copy of the License at
- #
- # http://www.apache.org/licenses/LICENSE-2.0
- #
- # Unless required by applicable law or agreed to in writing, software
- # distributed under the License is distributed on an "AS IS" BASIS,
- # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- # See the License for the specific language governing permissions and
- # limitations under the License.
- # ============================================================================
- """post process for 310 inference"""
-
- import sys
- import argparse
- import os.path as osp
- import numpy as np
-
- from src.utils import Logger
- from src.eval_metrics import evaluate
-
- parser = argparse.ArgumentParser(description='Train AlignedReID with cross entropy loss and triplet hard loss')
- parser.add_argument('--q_feature', type=str, default='')
- parser.add_argument('--q_localfeature', type=str, default='')
- parser.add_argument('--q_pid', type=str, default='')
- parser.add_argument('--q_camid', type=str, default='')
- parser.add_argument('--g_feature', type=str, default='')
- parser.add_argument('--g_localfeature', type=str, default='')
- parser.add_argument('--g_pid', type=str, default='')
- parser.add_argument('--g_camid', type=str, default='')
- parser.add_argument('--train_url', type=str, default='log')
- parser.add_argument('--reranking', type=lambda x: x.lower() == 'true', default=True, help='re_rank')
- parser.add_argument('--test_distance', type=str, default='global_local', help='test distance type')
- parser.add_argument('--unaligned', action='store_true')
- args = parser.parse_args()
-
- def get_query(feature_file, localfeature_file, pid_file, camid_file):
- """get query data"""
- qf, lqf, q_pids, q_camids = [], [], [], []
- openfilef = open(feature_file, 'r')
- for line in openfilef.readlines():
- temp = float(line)
- qf.append(temp)
- openfilel = open(localfeature_file, 'r')
- for line in openfilel.readlines():
- temp = float(line)
- lqf.append(temp)
- openfilep = open(pid_file, 'r')
- for line in openfilep.readlines():
- temp = int(line)
- q_pids.append(temp)
- openfilec = open(camid_file, 'r')
- for line in openfilec.readlines():
- temp = int(line)
- q_camids.append(temp)
- qf = np.array(qf)
- qf = qf.reshape(-1, 2048)
- lqf = np.array(lqf)
- lqf = lqf.reshape((-1, 2048, 8))
- q_pids = np.asarray(q_pids)
- q_camids = np.asarray(q_camids)
- print("Extracted features for query set, obtained {}-by-{} matrix".format(qf.shape[0], qf.shape[1]))
- return qf, lqf, q_pids, q_camids
-
- def get_gallery(feature_file, localfeature_file, pid_file, camid_file):
- """get gallery data"""
- gf, lgf, g_pids, g_camids = [], [], [], []
- openfilef = open(feature_file, 'r')
- for line in openfilef.readlines():
- temp = float(line)
- gf.append(temp)
- openfilel = open(localfeature_file, 'r')
- for line in openfilel.readlines():
- temp = float(line)
- lgf.append(temp)
- openfilep = open(pid_file, 'r')
- for line in openfilep.readlines():
- temp = int(line)
- g_pids.append(temp)
- openfilec = open(camid_file, 'r')
- for line in openfilec.readlines():
- temp = int(line)
- g_camids.append(temp)
- gf = np.array(gf)
- gf = gf.reshape(-1, 2048)
- lgf = np.array(lgf)
- lgf = lgf.reshape((-1, 2048, 8))
- g_pids = np.asarray(g_pids)
- g_camids = np.asarray(g_camids)
- print("Extracted features for gallery set, obtained {}-by-{} matrix".format(gf.shape[0], gf.shape[1]))
- return gf, lgf, g_pids, g_camids
-
- def re_ranking(probFea, galFea, k1, k2, lambda_value, local_distmat=None, only_local=False):
- """compute reranking when test"""
- query_num = probFea.shape[0]
- all_num = query_num + galFea.shape[0]
- if only_local:
- original_dist = local_distmat
- else:
- feat = np.concatenate([probFea, galFea])
- feat2 = feat
-
- feat = np.power(feat, 2)
- feat = np.sum(feat, axis=1, keepdims=True)
- feat = np.broadcast_to(feat, (all_num, all_num))
- feat_transpose = np.transpose(feat, (1, 0))
- distmat = feat+feat_transpose
-
- feat2_transpose = np.transpose(feat2, (1, 0))
- opt1 = np.dot(feat2, feat2_transpose)
- distmat = distmat + opt1*(-2)
-
- original_dist = distmat
- del feat
- if not local_distmat is None:
- original_dist = original_dist + local_distmat
- 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]
- leng1 = len(np.intersect1d(candidate_k_reciprocal_index, k_reciprocal_index))
- if leng1 > 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 = [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 test(qf, gf, lqf, lgf, q_pids, g_pids, q_camids, g_camids):
- """test"""
- ranks = [1, 5, 10, 20]
-
- m_qf = np.linalg.norm(x=qf, ord=2, axis=-1, keepdims=True)
- m_qf = np.broadcast_to(m_qf, qf.shape)
- qf = 1.*qf /(m_qf + 1e-12)
- m_gf = np.linalg.norm(x=gf, ord=2, axis=-1, keepdims=True)
- m_gf = np.broadcast_to(m_gf, gf.shape)
- gf = 1.*gf /(m_gf + 1e-12)
-
- m, n = qf.shape[0], gf.shape[0]
- distqf = np.power(qf, 2)
- distqf = np.sum(distqf, axis=1, keepdims=True)
- distqf = np.broadcast_to(distqf, (m, n))
- distgf = np.power(gf, 2)
- distgf = np.sum(distgf, axis=1, keepdims=True)
- distgf = np.broadcast_to(distgf, (n, m))
-
- transpose_distgf = np.transpose(distgf, (1, 0))
- distmat = distqf + transpose_distgf
- transpose_gf = np.transpose(gf, (1, 0))
- temp = np.matmul(qf, transpose_gf)
- temp = temp*(-2)
- distmat = distmat+temp # global_distmat
-
- if not args.test_distance == 'global':
- from src.distance import low_memory_local_dist
- lqf = np.transpose(lqf, (0, 2, 1))
- lgf = np.transpose(lgf, (0, 2, 1))
- local_distmat = low_memory_local_dist(lqf, lgf, aligned=not args.unaligned) # local_distmat
- if args.test_distance == 'local':
- distmat = local_distmat
- if args.test_distance == 'global_local':
- print("Using global and local branches")
- distmat = local_distmat+distmat
- print("Computing CMC and mAP")
- cmc, mAP = evaluate(distmat, q_pids, g_pids, q_camids, g_camids)
-
- print("Results ----------")
- print("mAP: {:.1%}".format(mAP))
- print("CMC curve")
- for r in ranks:
- print("Rank-{:<3}: {:.1%}".format(r, cmc[r - 1]))
-
- if args.reranking:
- if args.test_distance == 'global':
- print("Only using global branch for reranking")
- distmat = re_ranking(qf, gf, k1=20, k2=6, lambda_value=0.3)
- else:
- local_qq_distmat = low_memory_local_dist(lqf, lqf, aligned=not args.unaligned)
- local_gg_distmat = low_memory_local_dist(lgf, lgf, aligned=not args.unaligned)
- local_dist = np.concatenate(
- [np.concatenate([local_qq_distmat, local_distmat], axis=1),
- np.concatenate([local_distmat.T, local_gg_distmat], axis=1)],
- axis=0)
- if args.test_distance == 'local':
- print("Only using local branch for reranking")
- distmat = re_ranking(qf, gf, k1=20, k2=6, lambda_value=0.3, local_distmat=local_dist, only_local=True)
- elif args.test_distance == 'global_local':
- print("Using global and local branches for reranking")
- distmat = re_ranking(qf, gf, k1=20, k2=6, lambda_value=0.3, local_distmat=local_dist, only_local=False)
- print("Computing CMC and mAP for re_ranking")
- cmc, mAP = evaluate(distmat, q_pids, g_pids, q_camids, g_camids)
-
- print("Results ----------")
- print("mAP(RK): {:.1%}".format(mAP))
- print("CMC curve(RK)")
- for r in ranks:
- print("Rank-{:<3}: {:.1%}".format(r, cmc[r - 1]))
- return cmc[0]
-
- if __name__ == '__main__':
- sys.stdout = Logger(osp.join(args.train_url, 'log_test.txt'))
- sqf, slqf, sq_pids, sq_camids = get_query(args.q_feature, args.q_localfeature, args.q_pid, args.q_camid)
- sgf, slgf, sg_pids, sg_camids = get_gallery(args.g_feature, args.g_localfeature, args.g_pid, args.g_camid)
- test(sqf, sgf, slqf, slgf, sq_pids, sg_pids, sq_camids, sg_camids)
|