def print_ced_compare_methods(
method_errors,method_names,test_data,log_path='', save_log=True, norm='interocular distance'):
plt.yticks(np.linspace(0, 1, 11))
plot_cumulative_error_distribution(
[list(err) for err in list(method_errors)],
legend_entries=list(method_names),
marker_style=['s'],
marker_size=7,
x_label='Normalised Point-to-Point Error\n('+norm+')\n*'+test_data+' set*'
)
if save_log:
plt.savefig(os.path.join(log_path,'nme_ced_on_'+test_data+'_set.png'), bbox_inches='tight')
print ('ced plot path: ' + os.path.join(log_path,'nme_ced_on_'+test_data+'_set.png'))
plt.close()
def plot_ced_3dMDLab(errors):
errs = [[
error_vtx for error_mesh in errors[key] for error_vtx in error_mesh
] for key in errors.keys()]
legs = [method for method in errors.keys()]
tab = statistics_table(errs, legs, 0.105, 0.01, precision=4, sort_by='auc')
r = plot_cumulative_error_distribution(
errs,
legend_font_size=20,
line_width=5,
marker_size=10,
axes_font_size=20,
error_range=[0, 0.105, 0.01],
marker_edge_width=4,
y_label='Vertexes proportion',
x_label='Normalized dense vertex error',
title='',
legend_entries=legs,
axes_y_ticks=list(np.arange(0., 1.05, 0.1)),
figure_size=(7, 5))
plt.gca().set_axisbelow(True)
plt.gca().yaxis.grid(color='gray', linestyle='dashed')
return tab, r
def plot_ced(errors, method_names=['MDM']):
from menpofit.visualize import plot_cumulative_error_distribution
# plot the ced and store it at the root.
fig = plt.figure()
fig.add_subplot(111)
plot_cumulative_error_distribution(errors, legend_entries=method_names,
error_range=(0, 0.09, 0.005))
# shift the main graph to make room for the legend
ax = plt.gca()
box = ax.get_position()
ax.set_position([box.x0, box.y0, box.width * 0.9, box.height])
fig.canvas.draw()
data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')
data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
plt.clf()
return data
def plot_ced(errors, method_names=['MDM']):
from matplotlib import pyplot as plt
from menpofit.visualize import plot_cumulative_error_distribution
import numpy as np
# plot the ced and store it at the root.
fig = plt.figure()
fig.add_subplot(111)
plot_cumulative_error_distribution(errors, legend_entries=method_names,
error_range=(0, 0.09, 0.005))
# shift the main graph to make room for the legend
ax = plt.gca()
box = ax.get_position()
ax.set_position([box.x0, box.y0, box.width * 0.9, box.height])
fig.canvas.draw()
data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')
data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
plt.clf()
return data
def print_nme_statistics(
errors, model_path, network_type, test_data, max_error=0.08, log_path='', save_log=True, plot_ced=True,
norm='interocular distance'):
auc, failures = AUC(errors, max_error=max_error)
print ("\n****** NME statistics for " + network_type + " Network ******\n")
print ("* model path: " + model_path)
print ("* dataset: " + test_data + ' set')
print ("\n* Normalized mean error (percentage of "+norm+"): %.2f" % (100 * np.mean(errors)))
print ("\n* AUC @ %.2f: %.2f" % (max_error, 100 * auc))
print ("\n* failure rate @ %.2f: %.2f" % (max_error, 100 * failures) + '%')
if plot_ced:
plt.figure()
plt.yticks(np.linspace(0, 1, 11))
plot_cumulative_error_distribution(
list(errors),
legend_entries=[network_type],
marker_style=['s'],
marker_size=7,
x_label='Normalised Point-to-Point Error\n('+norm+')\n*' + test_data + ' set*',
)
if save_log:
with open(os.path.join(log_path, network_type.lower() + "_nme_statistics_on_" + test_data + "_set.txt"),
"wb") as f:
f.write(b"************************************************")
f.write(("\n****** NME statistics for " + str(network_type) + " Network ******\n").encode())
f.write(b"************************************************")
f.write(("\n\n* model path: " + str(model_path)).encode())
f.write(("\n\n* dataset: " + str(test_data) + ' set').encode())
f.write(b"\n\n* Normalized mean error (percentage of "+norm+"): %.2f" % (100 * np.mean(errors)))
f.write(b"\n\n* AUC @ %.2f: %.2f" % (max_error, 100 * auc))
f.write(("\n\n* failure rate @ %.2f: %.2f" % (max_error, 100 * failures) + '%').encode())
if plot_ced:
plt.savefig(os.path.join(log_path, network_type.lower() + '_nme_ced_on_' + test_data + '_set.png'),
bbox_inches='tight')
plt.close()
print ('\nlog path: ' + log_path)