def runSMCNTF(dbFeat, qFeat, gt, SMCN_params, err, flag=None):
st = time.time()
n1 = np.linalg.norm(dbFeat, axis=1).reshape((-1, 1))
n2 = np.linalg.norm(qFeat, axis=1).reshape((1, -1))
n12 = np.matmul(n1, n2)
S_pairwise = np.matmul(dbFeat, qFeat.T) / n12
S_in = np.matmul(dbFeat, dbFeat.T) / np.matmul(n1, n1.T)
can = getCandidate(S_in, S_pairwise, *SMCN_params)
S_smcntf, pathid = refreshLoss(S_in, S_pairwise, can, err)
time_cost = time.time() - st
if flag == 's':
return S_smcntf
elif flag == 'r':
return pathid
else:
P, R = utils.drawPR(S_smcntf, gt, True)
ap = utils.calAvgPred(P, R)
return ap, time_cost
def runSMCN(dbFeat, qFeat, gt, SMCN_params, flag=None):
st = time.time()
n1 = np.linalg.norm(dbFeat, axis=1).reshape((-1, 1))
n2 = np.linalg.norm(qFeat, axis=1).reshape((1, -1))
n12 = np.matmul(n1, n2)
S_pairwise = np.matmul(dbFeat, qFeat.T) / n12
S_in = np.matmul(dbFeat, dbFeat.T) / np.matmul(n1, n1.T)
can = getCandidate(S_in, S_pairwise, *SMCN_params)
S_smcncan = np.zeros_like(S_pairwise)
for i in range(can.shape[1]):
tmp = np.where(can[:, i] >= 0)[0]
if len(tmp):
S_smcncan[can[tmp, i], i] = S_pairwise[can[tmp, i], i]
if flag == 's':
return S_smcncan
else:
time_cost = time.time() - st
P, R = utils.drawPR(S_smcncan, gt)
ap = utils.calAvgPred(P, R)
return ap, time_cost
subdir = arg_dataset['subdir']
db_save_path = join(desc_dir, subdir[arg_eval['compare_subdir'][0]] + '_desc.npy')
q_save_path = join(desc_dir, subdir[arg_eval['compare_subdir'][1]] + '_desc.npy')
dbFeat = np.load(db_save_path)
qFeat = np.load(q_save_path)
if arg_dataset['name'] == 'scut':
dbFeat = dbFeat[::4, :]
qFeat = qFeat[::4, :]
dbn = dbFeat.shape[0]
qn = qFeat.shape[0]
# get ground truth
if arg_dataset['name'] == 'ox_robotcar':
err = arg_dataset['err']
db_gps = np.load(join(imgdir, 'gps_' + subdir[arg_eval['compare_subdir'][0]] + '.npy'))
q_gps = np.load(join(imgdir, 'gps_' + subdir[arg_eval['compare_subdir'][1]] + '.npy'))
_, gtm = utils.getGpsGT(db_gps, q_gps, err)
else:
err = arg_dataset['err']
gtm = utils.getGroundTruthMatrix(dbn, err)
print("Computing Descriptors...")
seqLen = 16
descData = performDelta([dbFeat, qFeat], seqLen)
S = utils.MCN_pairwise(descData[0], descData[1])
P, R = utils.drawPR(S, gtm)
auc = utils.calAvgPred(P, R)
print("delta auc : %.4f" % auc)
sig_D1_slsbh = utils.getLSBH(sig_D1, P, 0.25) sig_D2_slsbh = utils.getLSBH(sig_D2, P, 0.25) del P # pairwise n1 = np.linalg.norm(sig_D1, axis=1).reshape((-1, 1)) n2 = np.linalg.norm(sig_D2, axis=1).reshape((1, -1)) n12 = np.matmul(n1, n2) S_pairwise = np.matmul(sig_D1, sig_D2.T) / n12 # sLBSH nOnes = 2 * s * new_dims S_slbsh = np.matmul(sig_D1_slsbh, sig_D2_slsbh.T, dtype=np.float) / nOnes P, R = utils.drawPR(S_pairwise, GT) ap_pairwise = utils.calAvgPred(P, R) P, R = utils.drawPR(S_slbsh, GT) ap_slsbh = utils.calAvgPred(P, R) # calulate the result and save them, firstly uncomment following code and # run them, and comment it and visualize the result # old_dims = omni_D1.shape[1] # newr = np.array([0.01, 0.02, 0.03, 0.04, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5]) #(*) # new_dims = (8192 * newr).astype(np.int) # s = 0.25 # # omni_ap = [] # for each_dim in new_dims: # P = np.random.rand(old_dims, each_dim) # P /= np.linalg.norm(P, axis=1, keepdims=True)
err = 9 #(*) error tolerance for nordland dataset, 9 is set in paper
S_file = '../cp3_5/vis3/nd_sumwin.npz' #(*) NL similarity matrix
S = np.load(S_file)
S_pw = S['S_pw']
S_pwk = S['S_pwk']
S_mcn = S['S_mcn']
S_seq = loadmat('../cp3_4/seqslam/nd_sumwin.mat')['S'] #(*)
tmp = np.isnan(S_seq)
S_seq[tmp] = np.max(S_seq[~tmp])
S_smcn = S['S_smcn']
S_smcntf = S['S_smcntf']
S_dd = S_nd_dd
del S
GT = utils.getGroundTruthMatrix(S_pw.shape[0], err)
P, R = utils.drawPR(S_pw, GT)
ap = utils.calAvgPred(P, R)
plt.plot(R, P, color=vis.color[0], label='PW: %.4f' % ap, linewidth=2)
P, R = utils.drawPR(S_pwk, GT)
ap = utils.calAvgPred(P, R)
plt.plot(R, P, color=vis.color[1], label='PWk: %.4f' % ap, linewidth=2)
P, R = utils.drawPR(S_mcn, GT)
ap = utils.calAvgPred(P, R)
plt.plot(R, P, color=vis.color[2], label='MCN: %.4f' % ap, linewidth=2)
P, R = utils.drawPR(S_seq, GT, True)
ap = utils.calAvgPred(P, R)
plt.plot(R, P, color=vis.color[3], label='SeqSLAM: %.4f' % ap, linewidth=2)
gt = utils.getGroundTruth(dbn, err)
gtm = utils.getGroundTruthMatrix(dbn, err)
# oxford robotcar
# err = arg_dataset['err']
# db_gps = np.load(join(imgdir, 'gps_' + subdir[arg_eval['compare_subdir'][0]] + '.npy'))
# q_gps = np.load(join(imgdir, 'gps_' + subdir[arg_eval['compare_subdir'][1]] + '.npy'))
# _, gtm = utils.getGpsGT(db_gps, q_gps, err)
S = utils.MCN_pairwise(dbFeat, qFeat)
# experiment 3_2
# if SCUT, uncomment following
# canr = [1, 3, 5, 7]
# otherwise
canr = [0.01, 0.02, 0.05, 0.1] #(*)
spid = np.argsort(S, axis=0)
for each in canr:
# if not SCUT #(*)
cannum = int(each * dbn)
# otherwise
# cannum = each
S_can = np.zeros_like(S)
for i in range(qn):
S_can[spid[-cannum:, i], i] = S[spid[-cannum:, i], i]
P, R = utils.drawPR(S_can, gtm)
auc = utils.calAvgPred(P, R)
print('cannum: %d auc: %.4f' % (cannum, auc))
S_omni_mcn = MCN.runMCN_SDR(params, omni_D1_slsbh, omni_D2_slsbh, GT)
# SMCN
can = SMCN.getCandidate(S_omni_in, S_omni, *(2, 1, 3))
S_omni_smcncan = np.zeros_like(S_omni)
for i in range(can.shape[1]):
tmp = np.where(can[:, i] >= 0)[0]
if len(tmp):
S_omni_smcncan[can[tmp, i], i] = S_omni[can[tmp, i], i]
# SMCNTF
S_omni_smcntf = SMCN.refreshLoss(S_omni_in, S_omni, can, 1)
# visulize
plt.figure()
P, R = utils.drawPR(S_pairwise, GT)
ap = utils.calAvgPred(P, R)
plt.plot(R, P, color=utils.color[0], label='pairwise: %.4f' % ap, linewidth=2)
P, R = utils.drawPR(S_slbsh, GT)
ap = utils.calAvgPred(P, R)
plt.plot(R, P, color=utils.color[1], label='sLBSH: %.4f' % ap, linewidth=2)
P, R = utils.drawPR(S_omni, GT)
ap = utils.calAvgPred(P, R)
plt.plot(R,
P,
color=utils.color[2],
label='omni-sLBSH%d : %.4f' % (new_dims * 2, ap),
linewidth=2)