def main(argv):
file = open(argv[0], 'r')
net = pickle.load(file)
print 'Loaded net'
isVal = len(argv) > 1 and argv[1] == 'val'
print 'isVal: ', isVal
X, y_gt = load2d(isVal=isVal, isTiny=False)
print 'Loaded data'
y_pred = net.predict(X)
print 'Predicted data'
pdj = getPDJ(y_gt, y_pred)
fig = pyplot.figure(figsize=(6, 6))
axes = pyplot.gca()
pyplot.grid()
axes.set_ylim([0.0, 1.0])
axes.set_xlim([0.0, 0.5])
pyplot.xlabel('Normalized distance to true joint')
pyplot.ylabel('Detection rate')
pyplot.title('PDJ for different limbs over various thresholds')
fractions = [0.0, 0.1, 0.2, 0.3, 0.4, 0.5]
pyplot.plot(fractions, pdj[:,0], linewidth=2, label='head', color='#FC474C')
pyplot.plot(fractions, pdj[:,1], linewidth=2, label='torso', color='#8DE047')
pyplot.plot(fractions, pdj[:,2], linewidth=2, label='arms', color='#FFDD50')
pyplot.plot(fractions, pdj[:,3], linewidth=2, label='legs', color='#53A3D7')
pyplot.legend(loc='upper left', shadow=True, fontsize='medium')
pyplot.savefig('detections')
def main(argv):
file = open(argv[0], 'r')
net = pickle.load(file)
print 'Loaded net'
isTest = len(argv) > 1 and argv[1] == 'test'
print 'isTest: ', isTest
X, _ = load2d(test=isTest, isTiny=False)
print 'Loaded data'
print 'X shape: ', X.shape
y_pred = net.predict(X)
print "y_pred shape: ", y_pred.shape
print "x shape: ", X.shape
print 'Predicted data'
fig = pyplot.figure(figsize=(6, 6))
fig.subplots_adjust(left=0, right=1, bottom=0, top=1, hspace=0.05, wspace=0.05)
for i in range(42):
ax = fig.add_subplot(6, 7, i + 1, xticks=[], yticks=[])
plot_sample(X[i], y_pred[i], ax)
pyplot.savefig('segments')
def submission(fname_net="net4.pickle"):
with open(fname_net, "rb") as f:
net = pickle.load(f) # net = specialists
X = load.load2d(test=True)[0]
y_pred = net.predict(X)
print "Finish predict test file"
columns = "left_eye_center_x,left_eye_center_y,right_eye_center_x,right_eye_center_y,left_eye_inner_corner_x,left_eye_inner_corner_y,left_eye_outer_corner_x,left_eye_outer_corner_y,right_eye_inner_corner_x,right_eye_inner_corner_y,right_eye_outer_corner_x,right_eye_outer_corner_y,left_eyebrow_inner_end_x,left_eyebrow_inner_end_y,left_eyebrow_outer_end_x,left_eyebrow_outer_end_y,right_eyebrow_inner_end_x,right_eyebrow_inner_end_y,right_eyebrow_outer_end_x,right_eyebrow_outer_end_y,nose_tip_x,nose_tip_y,mouth_left_corner_x,mouth_left_corner_y,mouth_right_corner_x,mouth_right_corner_y,mouth_center_top_lip_x,mouth_center_top_lip_y,mouth_center_bottom_lip_x,mouth_center_bottom_lip_y"
columns = columns.split(",")
y_pred = y_pred * 48 + 48
y_pred = y_pred.clip(0, 96)
df = DataFrame(y_pred, columns=columns)
lookup_table = read_csv(os.path.expanduser("./data/IdLookupTable.csv"))
values = []
for index, row in lookup_table.iterrows():
values.append((row["RowId"], df.ix[row.ImageId - 1][row.FeatureName]))
now_str = datetime.now().isoformat().replace(":", "-")
submission = DataFrame(values, columns=("RowId", "Location"))
filename = "submission-{}.csv".format(now_str)
submission.to_csv(filename, index=False)
print ("Wrote {}".format(filename))
def main(argv):
file = open(argv[0], "r")
net = pickle.load(file)
print "Loaded net"
isVal = len(argv) > 1 and argv[1] == "val"
print "isVal: ", isVal
X, y_gt = load2d(isVal=isVal, isTiny=False)
print "Loaded data"
y_pred = net.predict(X)
print "Predicted data"
pdj = getPDJ(y_gt, y_pred)
fig = pyplot.figure(figsize=(6, 6))
axes = pyplot.gca()
pyplot.grid()
axes.set_ylim([0.0, 1.0])
axes.set_xlim([0.0, 0.5])
pyplot.xlabel("Normalized distance to true joint")
pyplot.ylabel("Detection rate")
pyplot.title("PDJ for different limbs over various thresholds")
fractions = [0.0, 0.1, 0.2, 0.3, 0.4, 0.5]
pyplot.plot(fractions, pdj[:, 0], linewidth=2, label="head", color="#FC474C")
pyplot.plot(fractions, pdj[:, 1], linewidth=2, label="torso", color="#8DE047")
pyplot.plot(fractions, pdj[:, 2], linewidth=2, label="arms", color="#FFDD50")
pyplot.plot(fractions, pdj[:, 3], linewidth=2, label="legs", color="#53A3D7")
pyplot.legend(loc="upper left", shadow=True, fontsize="medium")
pyplot.savefig("detections")
def submission(fname_net='net4.pickle'):
with open(fname_net, 'rb') as f:
net = pickle.load(f) # net = specialists
X = load.load2d(test=True)[0]
y_pred = net.predict(X)
print 'Finish predict test file'
columns = 'left_eye_center_x,left_eye_center_y,right_eye_center_x,right_eye_center_y,left_eye_inner_corner_x,left_eye_inner_corner_y,left_eye_outer_corner_x,left_eye_outer_corner_y,right_eye_inner_corner_x,right_eye_inner_corner_y,right_eye_outer_corner_x,right_eye_outer_corner_y,left_eyebrow_inner_end_x,left_eyebrow_inner_end_y,left_eyebrow_outer_end_x,left_eyebrow_outer_end_y,right_eyebrow_inner_end_x,right_eyebrow_inner_end_y,right_eyebrow_outer_end_x,right_eyebrow_outer_end_y,nose_tip_x,nose_tip_y,mouth_left_corner_x,mouth_left_corner_y,mouth_right_corner_x,mouth_right_corner_y,mouth_center_top_lip_x,mouth_center_top_lip_y,mouth_center_bottom_lip_x,mouth_center_bottom_lip_y'
columns = columns.split(',')
y_pred = y_pred * 48 + 48
y_pred = y_pred.clip(0, 96)
df = DataFrame(y_pred, columns=columns)
lookup_table = read_csv(os.path.expanduser('./data/IdLookupTable.csv'))
values = []
for index, row in lookup_table.iterrows():
values.append((
row['RowId'],
df.ix[row.ImageId - 1][row.FeatureName],
))
now_str = datetime.now().isoformat().replace(':', '-')
submission = DataFrame(values, columns=('RowId', 'Location'))
filename = 'submission-{}.csv'.format(now_str)
submission.to_csv(filename, index=False)
print("Wrote {}".format(filename))
def main(argv):
file = open(argv[0], 'r')
net = pickle.load(file)
print 'Loaded net'
isTest = len(argv) > 1 and argv[1] == 'test'
print 'isTest: ', isTest
X, _ = load2d(test=isTest, isTiny=False)
print 'Loaded data'
print 'X shape: ', X.shape
y_pred = net.predict(X)
print "y_pred shape: ", y_pred.shape
print "x shape: ", X.shape
print 'Predicted data'
fig = pyplot.figure(figsize=(6, 6))
fig.subplots_adjust(left=0,
right=1,
bottom=0,
top=1,
hspace=0.05,
wspace=0.05)
for i in range(42):
ax = fig.add_subplot(6, 7, i + 1, xticks=[], yticks=[])
plot_sample(X[i], y_pred[i], ax)
pyplot.savefig('segments')
def main(argv):
file = open(argv[0], 'r')
net = pickle.load(file)
print 'Loaded net'
isTest = len(argv) > 1 and argv[1] == 'test'
print 'isTest: ', isTest
X, _ = load2d(test=isTest, isTiny=False)
print 'Loaded data'
y_pred = net.predict(X)
print "y_pred shape: ", y_pred.shape
print "X shape: ", X.shape
print 'Predicted data'
for i in range(y_pred.shape[0]):
print i
fig = pyplot.figure(figsize=(6, 6))
fig.subplots_adjust(left=0,
right=1,
bottom=0,
top=1,
hspace=0.05,
wspace=0.05)
plot_sample(X[i], y_pred[i])
pyplot.savefig('segments/' + str(i))
pyplot.close(fig)
def main(argv):
file = open(argv[0], 'r')
fraction = float(argv[1])
net = pickle.load(file)
isVal = (len(argv) > 2) and (argv[2] == 'val')
X, y_gt = load2d(isVal=isVal) # for training
y_pred = net.predict(X)
# 0-based indices for torso
shoulder_l = 9
hip_r = 2
num_examples = y_gt.shape[0]
num_joint_coords = y_gt.shape[1]
pdj = [0] * 14
# Iterates over each example
for i in xrange(num_examples):
num_correct_joints = 0
y_gt_tup = []
y_pred_tup = []
# Builds tuple vector of joints (x, y)
for j in xrange(0, num_joint_coords, 2):
y_gt_tup.append((y_gt[i, j], y_gt[i, j + 1]))
y_pred_tup.append((y_pred[i, j], y_pred[i, j + 1]))
torso_diameter = tup_distance(y_gt_tup[shoulder_l], y_gt_tup[hip_r])
error_radius = fraction * torso_diameter
# Iterates over each joint
for j in xrange(len(y_gt_tup)):
if withinThreshold(y_gt_tup[j], y_pred_tup[j], error_radius):
pdj[j] += 1
print "num examples: ", num_examples
pdj = [x / float(num_examples) for x in pdj]
print "PDJ: ", pdj
overall = sum(pdj) / len(pdj)
print "PDJ: ", overall
head = (pdj[12] + pdj[13]) / 2
torso = (pdj[12] + pdj[3] + pdj[2] + pdj[8] + pdj[9]) / 5
arms = (pdj[11] + pdj[10] + pdj[9] + pdj[8] + pdj[7] + pdj[6]) / 6
legs = (pdj[5] + pdj[4] + pdj[3] + pdj[2] + pdj[1] + pdj[0]) / 6
print "PDJ head, torso, arms, legs: ", [head, torso, arms, legs]
def main(argv):
file = open(argv[0], 'r')
net = pickle.load(file)
isTest = len(argv) > 1 and argv[1] == 'test'
X, _ = load2d(test=isTest)
y_pred = net.predict(X)
fig = pyplot.figure(figsize=(6, 6))
fig.subplots_adjust(left=0, right=1, bottom=0, top=1, hspace=0.05, wspace=0.05)
for i in range(16):
ax = fig.add_subplot(4, 4, i + 1, xticks=[], yticks=[])
plot_sample(X[i], y_pred[i], ax)
pyplot.savefig('samples')
def main(argv):
file = open(argv[0], 'r')
net = pickle.load(file)
isVal = (len(argv) > 2) and (argv[2] == 'val')
X, y_gt = load2d(isVal=isVal) # for training
print "y_gt shape: ", y_gt.shape
y_pred = net.predict(X)
total_limbs = 14 # 2x ankle-knee, 2x knee-hip, hip-hip, 2x hip-shoulder, neck-head, 2x shoulder-neck, 2x shoulder-elbow, 2x elbow-wrist
pair_indices = [(5, 4), (0, 1), (3, 4), (1, 2), (2, 3), (9, 3), (8, 2),
(10, 11), (9, 10), (9, 12), (12, 8), (7, 8), (6, 7),
(12, 13)]
#pair_indices = [(0, 1), (1, 2), (4, 5), (3, 4), (6, 7), (7, 8), (10, 11), (9, 10), (12, 13)]
#pair_indices = [(9, 12), (8, 12), (9, 3), (2, 3), (2, 8)] # torso
pcp = [0] * len(pair_indices)
num_examples = y_gt.shape[0]
num_joints = y_gt.shape[1]
# Iterates over each example
for i in xrange(num_examples):
y_gt_tup = []
y_pred_tup = []
# Builds tuple vector of joints (x, y)
for j in xrange(0, num_joints, 2):
y_gt_tup.append((y_gt[i, j], y_gt[i, j + 1]))
y_pred_tup.append((y_pred[i, j], y_pred[i, j + 1]))
# Iterates over each joint pairing
#for a,b in pair_indices:
for x in range(len(pair_indices)):
a, b = pair_indices[x]
gt_distance = tup_distance(y_gt_tup[a], y_gt_tup[b])
error_radius = 0.5 * gt_distance
if withinThreshold(y_pred_tup[a], y_gt_tup[a],
error_radius) or withinThreshold(
y_pred_tup[b], y_gt_tup[b], error_radius):
pcp[x] += 1
pcp = [x / float(num_examples) for x in pcp]
print "PCP: ", pcp
def main(argv): file = open(argv[0], 'r') net = pickle.load(file) isVal = (len(argv) > 2) and (argv[2] == 'val') X, y_gt = load2d(isVal=isVal) # for training print "y_gt shape: ", y_gt.shape y_pred = net.predict(X) total_limbs = 14 # 2x ankle-knee, 2x knee-hip, hip-hip, 2x hip-shoulder, neck-head, 2x shoulder-neck, 2x shoulder-elbow, 2x elbow-wrist pair_indices = [(5, 4), (0, 1), (3, 4), (1, 2), (2, 3), (9, 3), (8, 2), (10, 11), (9, 10), (9, 12), (12, 8), (7, 8), (6, 7), (12, 13)] #pair_indices = [(0, 1), (1, 2), (4, 5), (3, 4), (6, 7), (7, 8), (10, 11), (9, 10), (12, 13)] #pair_indices = [(9, 12), (8, 12), (9, 3), (2, 3), (2, 8)] # torso pcp = [0] * len(pair_indices) num_examples = y_gt.shape[0] num_joints = y_gt.shape[1] # Iterates over each example for i in xrange(num_examples): y_gt_tup = [] y_pred_tup = [] # Builds tuple vector of joints (x, y) for j in xrange(0, num_joints, 2): y_gt_tup.append((y_gt[i, j], y_gt[i, j+1])) y_pred_tup.append((y_pred[i, j], y_pred[i, j+1])) # Iterates over each joint pairing #for a,b in pair_indices: for x in range(len(pair_indices)): a, b = pair_indices[x] gt_distance = tup_distance(y_gt_tup[a], y_gt_tup[b]) error_radius = 0.5 * gt_distance if withinThreshold(y_pred_tup[a], y_gt_tup[a], error_radius) or withinThreshold(y_pred_tup[b], y_gt_tup[b], error_radius): pcp[x] += 1 pcp = [x / float(num_examples) for x in pcp] print "PCP: ", pcp
def main(argv):
file = open(argv[0], 'r')
net = pickle.load(file)
isTest = len(argv) > 1 and argv[1] == 'test'
X, _ = load2d(test=isTest)
y_pred = net.predict(X)
fig = pyplot.figure(figsize=(6, 6))
fig.subplots_adjust(left=0,
right=1,
bottom=0,
top=1,
hspace=0.05,
wspace=0.05)
for i in range(16):
ax = fig.add_subplot(4, 4, i + 1, xticks=[], yticks=[])
plot_sample(X[i], y_pred[i], ax)
pyplot.savefig('samples')
def main():
csv.field_size_limit(sys.maxsize)
sys.setrecursionlimit(10000)
X, y = load2d() # load 2-d data
print "Loaded data"
#net_pretrain = None
#with open('kfkd-net3.pickle', 'rb') as f:
# net_pretrain = pickle.load(f)
#from sklearn.base import clone
#net3 = clone(net_pretrain)
#set number of epochs relative to number of training examples:
#net3.max_epochs = int(1e7 / y.shape[0])
net3.fit(X, y)
with open('net3.pickle', 'wb') as f:
pickle.dump(net3, f, -1)
print "MSE: ", mean_squared_error(net3.predict(X), y)
def main():
csv.field_size_limit(sys.maxsize)
sys.setrecursionlimit(10000)
X, y = load2d() # load 2-d data
print "Loaded data"
#net_pretrain = None
#with open('kfkd-net3.pickle', 'rb') as f:
# net_pretrain = pickle.load(f)
#from sklearn.base import clone
#net3 = clone(net_pretrain)
#set number of epochs relative to number of training examples:
#net3.max_epochs = int(1e7 / y.shape[0])
net3.fit(X, y)
with open('net3.pickle', 'wb') as f:
pickle.dump(net3, f, -1)
print "MSE: ", mean_squared_error(net3.predict(X), y)
def main(argv):
file = open(argv[0], 'r')
net = pickle.load(file)
print 'Loaded net'
isTest = len(argv) > 1 and argv[1] == 'test'
print 'isTest: ', isTest
X, _ = load2d(test=isTest, isTiny=False)
print 'Loaded data'
y_pred = net.predict(X)
print "y_pred shape: ", y_pred.shape
print "X shape: ", X.shape
print 'Predicted data'
for i in range(y_pred.shape[0]):
print i
fig = pyplot.figure(figsize=(6, 6))
fig.subplots_adjust(left=0, right=1, bottom=0, top=1, hspace=0.05, wspace=0.05)
plot_sample(X[i], y_pred[i])
pyplot.savefig('segments/' + str(i))
pyplot.close(fig)
import cPickle as pickle
with open('SpeM2.pickle', 'wb') as f4:
pickle.dump(net_spe_M, f4, -1)
########################################################
# load the full net and the specialists sequentially #
########################################################
# ------ make full image prediction first ----------- #
with open('net4.pickle', 'rb') as f0:
net_full = pickle.load(f0)
X = load.load2d(test=True)[0]
print X.shape
y_pred = net_full.predict(X)
y_pred_final1 = np.zeros(y_pred.shape) # array to save the final predictions of each specialists
y_pred_final2 = np.zeros(y_pred.shape)
# ------ load the trained nets together to make the final predict ---------- #
# LE
# with open('SpeLE.pickle', 'rb') as f2:
# net_spe_LE = pickle.load(f2)
X_LE_test, x_co_LE, y_co_LE = load.load_spe_test(win=win, X_test=X, Y_pred=y_pred, indices=indices_LE)
y_LE_pred = net_spe_LE.predict(X_LE_test)
y_LE_pred_big = np.zeros(y_LE_pred.shape)
num_r = y_LE_pred.shape[0]
update_momentum=theano.shared(float32(0.9)),
regression=True,
batch_iterator_train=FlipBatchIterator2d(batch_size=128),
on_epoch_finished=[
AdjustVariable('update_learning_rate', start=0.03, stop=0.0001),
AdjustVariable('update_momentum', start=0.9, stop=0.999),
],
max_epochs=3000,
verbose=1,
)
import sys
sys.setrecursionlimit(10000)
X, y = load.load2d() # load 2-d data
net4.fit(X, y)
# Training for 1000 epochs will take a while. We'll pickle the
# trained model so that we can load it back later:
import cPickle as pickle
with open('net4.pickle', 'wb') as f:
pickle.dump(net4, f, -1)
# make submission
print 'start making submission'
def submission(fname_net='net4.pickle'):
with open(fname_net, 'rb') as f:
net = pickle.load(f) # net = specialists
update_momentum=theano.shared(float32(0.9)),
regression=True,
batch_iterator_train=FlipBatchIterator2d(batch_size=128),
on_epoch_finished=[
AdjustVariable("update_learning_rate", start=0.03, stop=0.0001),
AdjustVariable("update_momentum", start=0.9, stop=0.999),
],
max_epochs=3000,
verbose=1,
)
import sys
sys.setrecursionlimit(10000)
X, y = load.load2d() # load 2-d data
net4.fit(X, y)
# Training for 1000 epochs will take a while. We'll pickle the
# trained model so that we can load it back later:
import cPickle as pickle
with open("net4.pickle", "wb") as f:
pickle.dump(net4, f, -1)
# make submission
print "start making submission"
def submission(fname_net="net4.pickle"):
with open(fname_net, "rb") as f:
def MSE(argv): file = open(argv[0], 'r') net = pickle.load(file) X, y = load2d() # load 2-d data print "MSE: ", mean_squared_error(net.predict(X), y)
net_spe_M.fit(X_M, Y_M)
# pickle the trained specialists
import cPickle as pickle
with open('SpeM2.pickle', 'wb') as f4:
pickle.dump(net_spe_M, f4, -1)
########################################################
# load the full net and the specialists sequentially #
########################################################
# ------ make full image prediction first ----------- #
with open('net4.pickle', 'rb') as f0:
net_full = pickle.load(f0)
X = load.load2d(test=True)[0]
print X.shape
y_pred = net_full.predict(X)
y_pred_final1 = np.zeros(
y_pred.shape) # array to save the final predictions of each specialists
y_pred_final2 = np.zeros(y_pred.shape)
# ------ load the trained nets together to make the final predict ---------- #
# LE
# with open('SpeLE.pickle', 'rb') as f2:
# net_spe_LE = pickle.load(f2)
X_LE_test, x_co_LE, y_co_LE = load.load_spe_test(win=win,
X_test=X,
Y_pred=y_pred,
def MSE(argv): file = open(argv[0], 'r') net = pickle.load(file) X, y = load2d(isVal=True) # load 2-d data print "MSE: ", mean_squared_error(net.predict(X), y)