def pickle2npz():
nn12 = Cnnl('12-net').__load_model_old__('12-net_lasagne_.pickle')
nn12.save_model('12-net_lasagne_.npz')
nn24 = Cnnl(nn_name='24-net',
subnet=nn12).__load_model_old__('24-net_lasagne_.pickle')
nn48 = Cnnl(nn_name='48-net',
subnet=nn24).__load_model_old__('48-net_lasagne_.pickle')
nn24.save_model('24-net_lasagne_.npz')
nn48.save_model('48-net_lasagne_.npz')
def train(nn_name='12-net', k=12):
"""
Fucntion for traning 12-net with testing on part of data
using cross validation
"""
suff = str(k)
if nn_name.find('calib') > 0:
X_data_name = 'train_data_icalib_' + suff + '.npy'
y_data_name = 'labels_icalib_' + suff + '.npy'
else:
X_data_name = 'train_data_' + suff + '.npy'
y_data_name = 'labels_' + suff + '.npy'
rates12 = sp.hstack((0.05 * sp.ones(25, dtype=sp.float32),
0.005 * sp.ones(15, dtype=sp.float32),
0.0005 * sp.ones(10, dtype=sp.float32)))
rates24 = sp.hstack((0.01 * sp.ones(25, dtype=sp.float32),
0.0001 * sp.ones(15, dtype=sp.float32)))
rates48 = sp.hstack([
0.05 * sp.ones(15, dtype=sp.float32),
0.005 * sp.ones(10, dtype=sp.float32)
])
if nn_name == '24-net':
nn = Cnnl(
nn_name=nn_name,
l_rates=rates24,
subnet=Cnnl(nn_name='12-net',
l_rates=rates12).load_model('12-net_lasagne_.pickle'))
elif nn_name == '48-net':
nn = Cnnl(nn_name=nn_name,
l_rates=rates48,
subnet=Cnnl(
nn_name='24-net',
l_rates=rates24,
subnet=Cnnl(
nn_name='12-net',
l_rates=rates12).load_model('12-net_lasagne_.pickle')
).load_model('24-net_lasagne_.pickle'))
else:
nn = Cnnl(nn_name=nn_name, l_rates=rates12)
if not os.path.exists(nn_name + '_lasagne_.pickle'):
if nn_name.find('calib') > 0:
ds.get_train_wider_calib_data(k=k)
else:
ds.get_train_data(k=k)
X, y = sp.load(X_data_name), sp.load(y_data_name)
X_train, y_train = X, y
if not os.path.exists(nn_name + '_lasagne_.pickle'):
if nn_name == '24-net':
X_sub_train12 = sp.load('train_data_12.npy')
nn.fit(X=X_train, y=y_train, X12=X_sub_train12)
elif nn_name == '48-net':
X_sub_train12 = sp.load('train_data_12.npy')
X_sub_train24 = sp.load('train_data_24.npy')
nn.fit(X=X_train, y=y_train, X12=X_sub_train12, X24=X_sub_train24)
else:
nn.fit(X=X_train, y=y_train)
nn.save_model(nn_name + '_lasagne_.pickle')
def main():
iname = 'img/test/1226.jpg'
nn12 = Cnnl('12-net').load_model('12-net_lasagne_.pickle')
nn_calib12 = Cnnl('12-calib_net').load_model(
'12-calib_net_lasagne_.pickle')
nn24 = Cnnl(nn_name='24-net',
subnet=nn12).load_model('24-net_lasagne_.pickle')
nn_calib24 = Cnnl('24-calib_net').load_model(
'24-calib_net_lasagne_.pickle')
nn48 = Cnnl(nn_name='48-net',
subnet=nn24).load_model('48-net_lasagne_.pickle')
nn_calib48 = Cnnl('48-calib_net').load_model(
'48-calib_net_lasagne_.pickle')
image = fr.get_frame(iname)
p = 16
k = 1.18
bb = apply_12_net(image=image,
iname=iname,
nn=nn12,
p_level=p,
k=k,
debug=0)
bb = apply_calib_net(image=image, iname=iname, bb=bb, nn=nn_calib12)
bb = util.nms(bb, T=0.8)
bb = apply_24_net(image=image, iname=iname, bb=bb, nn=nn24)
bb = apply_calib_net(image=image, iname=iname, bb=bb, nn=nn_calib24)
bb = util.nms(bb, T=0.8)
bb = apply_48_net(image=image, iname=iname, bb=bb, nn=nn48)
bb = util.nms(bb, T=0.3)
bb = apply_calib_net(image=image, iname=iname, bb=bb, nn=nn_calib48)
draw_bboxes(iname, bb)
def train(nn_name='12-net', rootdir='F:\\'):
"""
Fucntion for traning 12-net with testing on part of data
using cross validation
"""
is_calib = nn_name.find('calib') > 0
X_name = 'train_data_' + ('icalib_'
if is_calib else '') + nn_name[:2] + '.npz'
y_name = 'labels_' + ('icalib_' if is_calib else '') + nn_name[:2] + '.npz'
is_calib = nn_name.find('calib') > 0
rates = sp.hstack((0.05 * sp.ones(333, dtype=sp.float32),
0.005 * sp.ones(333, dtype=sp.float32),
0.0005 * sp.ones(333, dtype=sp.float32),
0.00005 * sp.ones(5, dtype=sp.float32)))
rates12 = rates
rates24 = rates
rates48 = rates
rates12_c = rates12[333:666]
rates48_c = rates12_c
rates24_c = rates12_c
if nn_name == '24-net':
if is_calib:
nn = Cnnl(nn_name=nn_name, l_rates=rates24_c)
else:
nn = Cnnl(nn_name=nn_name, l_rates=rates24)
elif nn_name == '48-net':
if is_calib:
nn = Cnnl(nn_name=nn_name, l_rates=rates48_c)
else:
nn = Cnnl(nn_name=nn_name, l_rates=rates48)
else:
if is_calib:
nn = Cnnl(nn_name=nn_name, l_rates=rates12_c)
else:
nn = Cnnl(nn_name=nn_name, l_rates=rates12)
if is_calib:
X = util.load_from_npz(X_name)
y = util.load_from_npz(y_name)
nn.fit(X, y)
else:
rpath = os.path.join(rootdir, 'traindata', nn.nn_name)
nn.train_on_hdd(rootdir=rpath)
nn.__save_model_old__(nn_name + '_lasagne_.pickle')
def main():
iname = 'img/test/p1.jpg'
nn12 = Cnnl('12-net').__load_model_old__('12-net_lasagne_.pickle')
nn24 = Cnnl(nn_name = '24-net',subnet=nn12).__load_model_old__('24-net_lasagne_.pickle')
nn48 = Cnnl(nn_name = '48-net',subnet=nn24).__load_model_old__('48-net_lasagne_.pickle')
nn_calib12 = Cnnl('12-calib_net').__load_model_old__('12-calib_net_lasagne_.pickle')
nn_calib24 = Cnnl('24-calib_net').__load_model_old__('24-calib_net_lasagne_.pickle')
#nn_calib48 = Cnnl('48-calib_net').__load_model_old__('48-calib_net_lasagne_.pickle')
image = fr.get_frame(iname)
p = 11
k = 1.41
t_level = 0
gbb = []
gp = []
print('Start....')
s = time()
py = fr.get_frame_pyramid(image,scale=k,steps=p)
for m,frame in enumerate(py):
if m < t_level:
continue
bb,p = apply_12_net(image = image,frame=frame,iname = iname,nn = nn12,m=m,debug=0)
keep = util.nms(bb,p,T=0.5)
bb = bb[keep] if keep != [] else bb
p = p[keep] if keep != [] else p
cnt_detect = len(bb)
if cnt_detect == 0:
continue
#draw_bboxes(iname,bb,c='#-r_12--' + str(m))
for b in bb:
gbb.append(b)
for pp in p:
gp.append(pp)
gbb = sp.array(gbb)
gp = sp.array(gp)
#draw_bboxes(iname,gbb,c='0-r_12')
bb = gbb
p = gp
bb = apply_calib_net(image = image, iname = iname,bb=bb,nn=nn_calib12)
cnt_detect = len(bb)
#draw_bboxes(iname,bb,c='1-r_calib_12')
keep = util.nms(bb,p,T=0.3)
bb = bb[keep] if keep != [] else bb
p = p[keep] if keep != [] else p
#draw_bboxes(iname,bb,c='2-r_nms_12')
bb,p = apply_24_48_net(image = image, iname = iname, bb = bb,nn = nn24,debug=0 )
cnt_detect = len(bb)
#draw_bboxes(iname,bb,c='3-r_24')
bb = apply_calib_net(image = image, iname = iname,bb = bb,nn=nn_calib24)
cnt_detect = len(bb)
#draw_bboxes(iname,bb,c='4-r_calib_24')
keep = util.nms(bb,p,T=0.3)
bb = bb[keep] if keep != [] else bb
p = p[keep] if keep != [] else p
#draw_bboxes(iname,bb,c='5-r_nms_24')
cnt_detect = len(bb)
bb,p = apply_24_48_net(image = image, iname = iname, bb = bb,nn = nn48,debug=1)
cnt_detect = len(bb)
draw_bboxes(iname,bb,c='6-r_48',debug=1)
# bb = apply_calib_net(image = image,iname = iname,bb=bb,nn=nn_calib48)
# draw_bboxes(iname,bb,c='6-r_calib_48')
#
# bb = util.nms(bb,T=0.2)
# draw_bboxes(iname,bb,c='7-r_nms_48_final')
print('Finish...',time()-s)
def cv(nn_name, d_num=10000, k_fold=7, score_metrics='accuracy', verbose=0):
suff = str(nn_name[:2])
if nn_name.find('calib') > 0:
X_data_name = 'train_data_icalib_' + suff + '.npy'
y_data_name = 'labels_icalib_' + suff + '.npy'
else:
X_data_name = 'train_data_' + suff + '.npy'
y_data_name = 'labels_' + suff + '.npy'
X, y = sp.load(X_data_name), sp.load(y_data_name)
d_num = min(len(X), d_num)
X = X[:d_num]
y = y[:d_num]
rates12 = sp.hstack((0.05 * sp.ones(25, dtype=sp.float32),
0.005 * sp.ones(15, dtype=sp.float32),
0.0005 * sp.ones(10, dtype=sp.float32)))
rates24 = sp.hstack((0.01 * sp.ones(25, dtype=sp.float32),
0.0001 * sp.ones(15, dtype=sp.float32)))
rates48 = sp.hstack([
0.05 * sp.ones(15, dtype=sp.float32),
0.005 * sp.ones(10, dtype=sp.float32)
])
if nn_name == '48-net':
X12 = sp.load('train_data_12.npy')[:d_num]
X24 = sp.load('train_data_24.npy')[:d_num]
elif nn_name == '24-net':
X12 = sp.load('train_data_12.npy')[:d_num]
if score_metrics == 'accuracy':
score_fn = accuracy_score
else:
score_fn = f1_score
scores = []
iteration = 0
for t_indx, v_indx in util.kfold(X, y, k_fold=k_fold):
nn = None
X_train, X_test, y_train, y_test = X[t_indx], X[v_indx], y[t_indx], y[
v_indx]
#print('\t \t',str(iteration+1),'fold out of ',str(k_fold),'\t \t' )
if nn_name == '24-net':
nn = Cnnl(
nn_name=nn_name,
l_rates=rates24,
subnet=Cnnl(
nn_name='12-net',
l_rates=rates12).load_model('12-net_lasagne_.pickle'))
nn.fit(X=X_train, y=y_train, X12=X12[t_indx])
elif nn_name == '48-net':
nn = Cnnl(
nn_name=nn_name,
l_rates=rates48,
subnet=Cnnl(
nn_name='24-net',
l_rates=rates24,
subnet=Cnnl(
nn_name='12-net',
l_rates=rates12).load_model('12-net_lasagne_.pickle')).
load_model('24-net_lasagne_.pickle'))
nn.fit(X=X_train, y=y_train, X12=X12[t_indx], X24=X24[t_indx])
else:
nn = Cnnl(nn_name=nn_name, l_rates=rates12, verbose=verbose)
nn.fit(X=X_train, y=y_train)
if nn_name == '24-net':
y_pred = nn.predict(X_test, X12=X12[v_indx])
elif nn_name == '48-net':
y_pred = nn.predict(X_test, X12=X12[v_indx], X24=X24[v_indx])
else:
y_pred = nn.predict(X_test)
score = score_fn(y_test, y_pred)
#print(iteration,'fold score',score)
scores.append(score)
iteration += 1
score_mean = sp.array(scores).mean()
print(d_num, 'mean score', score)
return score_mean