mirrors
/
fast-reid
mirror of https://github.com/JDAI-CV/fast-reid.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

feat: support re-rank in test phase

pull/64/head
liaoxingyu 2020-05-13 11:47:52 +08:00
parent e502fadba9
commit 320010f2ae
4 changed files with 113 additions and 100 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
fastreid/config/defaults.py
View File

@ -222,7 +222,18 @@ _C.TEST = CN()
_C.TEST.EVAL_PERIOD = 50
_C.TEST.IMS_PER_BATCH = 128
# Precise BN
# Re-rank
_C.TEST.RERANK = CN()
_C.TEST.RERANK.ENABLED = False
_C.TEST.RERANK.K1 = 20
_C.TEST.RERANK.K2 = 6
_C.TEST.RERANK.LAMBDA = 0.3
# Average query expansion
_C.TEST.AQE = CN()
_C.TEST.AQE.ENABLED = True
# Precise batchnorm
_C.TEST.PRECISE_BN = CN()
_C.TEST.PRECISE_BN.ENABLED = False
_C.TEST.PRECISE_BN.DATASET = 'Market1501'

33
fastreid/evaluation/reid_evaluation.py
View File

@ -3,6 +3,7 @@
@author: liaoxingyu
@contact: sherlockliao01@gmail.com
"""
import logging
import copy
from collections import OrderedDict
@ -12,10 +13,14 @@ import torch.nn.functional as F
from .evaluator import DatasetEvaluator
from .rank import evaluate_rank
from .rerank import re_ranking
logger = logging.getLogger(__name__)
class ReidEvaluator(DatasetEvaluator):
def __init__(self, cfg, num_query, output_dir=None):
self.cfg = cfg
self._num_query = num_query
self._output_dir = output_dir
@ -29,14 +34,19 @@ class ReidEvaluator(DatasetEvaluator):
self.camids = []
def process(self, outputs):
self.features.append(outputs[0].cpu())
self.features.append(outputs[0])
self.pids.extend(outputs[1].cpu().numpy())
self.camids.extend(outputs[2].cpu().numpy())
@staticmethod
def cal_dist(query_feat: torch.tensor, gallery_feat: torch.tensor):
query_feat = F.normalize(query_feat, dim=1)
gallery_feat = F.normalize(gallery_feat, dim=1)
cos_dist = 1 - torch.mm(query_feat, gallery_feat.t()).cpu().numpy()
return cos_dist
def evaluate(self):
features = torch.cat(self.features, dim=0)
# normalize feature
features = F.normalize(features, dim=1)
# query feature, person ids and camera ids
query_features = features[:self._num_query]
@ -50,8 +60,21 @@ class ReidEvaluator(DatasetEvaluator):
self._results = OrderedDict()
cos_dist = torch.mm(query_features, gallery_features.t()).numpy()
cmc, all_AP, all_INP = evaluate_rank(1 - cos_dist, query_pids, gallery_pids, query_camids, gallery_camids)
dist = self.cal_dist(query_features, gallery_features)
if self.cfg.TEST.RERANK.ENABLED:
logger.info("Test with rerank setting")
k1 = self.cfg.TEST.RERANK.K1
k2 = self.cfg.TEST.RERANK.K1
lambda_value = self.cfg.TEST.RERANK.LAMBDA
q_q_dist = self.cal_dist(query_features, query_features)
g_g_dist = self.cal_dist(gallery_features, gallery_features)
dist = re_ranking(dist, q_q_dist, g_g_dist, k1, k2, lambda_value)
if self.cfg.TEST.AQE.ENABLED:
pass
cmc, all_AP, all_INP = evaluate_rank(dist, query_pids, gallery_pids, query_camids, gallery_camids)
mAP = np.mean(all_AP)
mINP = np.mean(all_INP)
for r in [1, 5, 10]:

73
fastreid/evaluation/rerank.py 100644
View File

@ -0,0 +1,73 @@
# encoding: utf-8
# based on:
# https://github.com/zhunzhong07/person-re-ranking
__all__ = ['re_ranking']
import numpy as np
def re_ranking(q_g_dist, q_q_dist, g_g_dist, k1: int = 20, k2: int = 6, lambda_value: float = 0.3):
original_dist = np.concatenate(
[np.concatenate([q_q_dist, q_g_dist], axis=1),
np.concatenate([q_g_dist.T, g_g_dist], axis=1)],
axis=0)
original_dist = np.power(original_dist, 2).astype(np.float32)
original_dist = np.transpose(1. * original_dist / np.max(original_dist, axis=0))
V = np.zeros_like(original_dist).astype(np.float32)
initial_rank = np.argsort(original_dist).astype(np.int32)
query_num = q_g_dist.shape[0]
gallery_num = q_g_dist.shape[0] + q_g_dist.shape[1]
all_num = gallery_num
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] = 1. * weight / np.sum(weight)
original_dist = original_dist[:query_num, ]
if k2 != 1:
V_qe = np.zeros_like(V, dtype=np.float32)
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.float32)
for i in range(query_num):
temp_min = np.zeros(shape=[1, gallery_num], dtype=np.float32)
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, V, jaccard_dist
final_dist = final_dist[:query_num, query_num:]
return final_dist

94
fastreid/utils/rerank.py
View File

@ -1,94 +0,0 @@
# encoding: utf-8
"""
Source: https://github.com/zhunzhong07/person-re-ranking
Created on Mon Jun 26 14:46:56 2017
@author: luohao
Modified by Houjing Huang, 2017-12-22.
- This version accepts distance matrix instead of raw features.
- The difference of `/` division between python 2 and 3 is handled.
- numpy.float16 is replaced by numpy.float32 for numerical precision.
CVPR2017 paper:Zhong Z, Zheng L, Cao D, et al. Re-ranking Person Re-identification with k-reciprocal Encoding[J]. 2017.
url:http://openaccess.thecvf.com/content_cvpr_2017/papers/Zhong_Re-Ranking_Person_Re-Identification_CVPR_2017_paper.pdf
Matlab version: https://github.com/zhunzhong07/person-re-ranking
API
q_g_dist: query-gallery distance matrix, numpy array, shape [num_query, num_gallery]
q_q_dist: query-query distance matrix, numpy array, shape [num_query, num_query]
g_g_dist: gallery-gallery distance matrix, numpy array, shape [num_gallery, num_gallery]
k1, k2, lambda_value: parameters, the original paper is (k1=20, k2=6, lambda_value=0.3)
Returns:
final_dist: re-ranked distance, numpy array, shape [num_query, num_gallery]
"""
__all__ = ['re_ranking']
import numpy as np
def re_ranking(q_g_dist, q_q_dist, g_g_dist, k1=20, k2=6, lambda_value=0.3):
# The following naming, e.g. gallery_num, is different from outer scope.
# Don't care about it.
original_dist = np.concatenate(
[np.concatenate([q_q_dist, q_g_dist], axis=1),
np.concatenate([q_g_dist.T, g_g_dist], axis=1)],
axis=0)
original_dist = np.power(original_dist, 2).astype(np.float32)
original_dist = np.transpose(1. * original_dist/np.max(original_dist,axis = 0))
V = np.zeros_like(original_dist).astype(np.float32)
initial_rank = np.argsort(original_dist).astype(np.int32)
query_num = q_g_dist.shape[0]
gallery_num = q_g_dist.shape[0] + q_g_dist.shape[1]
all_num = gallery_num
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] = 1.*weight/np.sum(weight)
original_dist = original_dist[:query_num,]
if k2 != 1:
V_qe = np.zeros_like(V,dtype=np.float32)
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.float32)
for i in range(query_num):
temp_min = np.zeros(shape=[1,gallery_num],dtype=np.float32)
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