-
Notifications
You must be signed in to change notification settings - Fork 0
/
results.py
340 lines (286 loc) · 13.2 KB
/
results.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
import json
import numpy as np
import os
from numpy import array as npa
import cv2
from utils_various import *
from PIL import Image
import matplotlib as mpl
from matplotlib import pyplot as plt
from matplotlib import colors # used for coloring bins in histograms
import matplotlib.patches as patches
from matplotlib.ticker import PercentFormatter # used for setting y-axis tick
# to percent of test images
from matplotlib.offsetbox import OffsetImage, AnnotationBbox, TextArea
from mpl_toolkits.axes_grid1.inset_locator import zoomed_inset_axes
from mpl_toolkits.axes_grid1.inset_locator import mark_inset
from cycler import cycler # used for changing default color sequence @ pyplot
plt.rc('text.latex', preamble=r'\usepackage{amsmath}'
r'\usepackage{amsfonts}')
mpl.rcParams['font.size'] = 12
mpl.rcParams['legend.fontsize'] = 'large'
mpl.rcParams['figure.titlesize'] = 'large'
mpl.rcParams['xtick.labelsize'] = 'medium'
mpl.rcParams['legend.framealpha'] = 0.65
mpl.rcParams['axes.labelsize'] = 'large'
def get_pose_error(r_GT, q_GT, r_pred, q_pred):
Et_vec = abs(r_GT - r_pred)
Et = np.linalg.norm(r_GT - r_pred, 2, axis=1)
et = Et/np.linalg.norm(r_GT, 2, axis=1)
eq = 2*np.arccos(np.minimum(1, np.abs(np.sum(q_GT*q_pred, axis=1)/(np.linalg.norm(q_GT, 2, axis=1)*np.linalg.norm(q_pred, 2, axis=1)))))
return Et_vec, Et, et, eq
def plot_wireframe(dcm, r, keypoints, bbox, ax, imgname, err_r, euler_angles, err_euler, line_color='cyan', line_width=1):
if ax is None:
ax = plt.gca()
# get landmarks corresponding to projection of predicted pose
xl, yl = project3Dto2D(dcm, r, Wireframe.landmark_mat)
img = cv2.imread("test/images/" + imgname)
rect = patches.Rectangle((bbox[0], bbox[1]), bbox[2], bbox[3], linewidth=1, edgecolor=(185/255,5/255,4/255), facecolor="none")
ax.add_patch(rect)
# bottom of main body
plt.plot(np.hstack((xl[0:4], xl[0])),
np.hstack((yl[0:4], yl[0])),
color=(108/255, 196/255, 193/255), lw=line_width)
plt.xlim([0, Camera.nu])
plt.ylim([Camera.nv, 0])
plt.xticks([])
plt.yticks([])
# solar panel
plt.plot(np.hstack((xl[4:8], xl[4])),
np.hstack((yl[4:8], yl[4])),
color=(108/255, 196/255, 193/255), lw=line_width)
# top of main body
x_top, y_top = project3Dto2D(dcm, r, Wireframe.topMainBody_mat)
plt.plot(np.hstack((x_top[0:4], x_top[0])),
np.hstack((y_top[0:4], y_top[0])),
color=(108/255, 196/255, 193/255), lw=line_width)
# corners
for k in np.arange(0, 4):
plt.plot(np.hstack((xl[k], x_top[k])),
np.hstack((yl[k], y_top[k])),
color=(108/255, 196/255, 193/255), lw=line_width)
# antennas
x_clamp, y_clamp = project3Dto2D(dcm, r, Wireframe.antClamps_mat)
for k in np.arange(0, 3):
plt.plot(np.hstack((x_clamp[k], xl[8 + k])),
np.hstack((y_clamp[k], yl[8 + k])),
color=(108/255, 196/255, 193/255), lw=line_width)
for i in range(11):
cv2.circle(img, (int(keypoints[i, 0]), int(keypoints[i, 1])), radius = 10, color = (108, 196, 193), thickness=-1)
ax.imshow(img)
box_propeties = dict(boxstyle='round', facecolor='gray', alpha=0.8, edgecolor='gray')
ax.text(0.03, 0.96,
'$\\bf{Distance}$ [m]\n\t$x$ = %.2f\n\t$y$ = %.2f\n\t$z$ = %.2f'
% (r[0], r[1], r[2])
+ '\n\nError [m]:\n\t$E_x = %.3f$\n\t$E_y = %.3f$\n\t$E_z = %.3f$'
% (err_r[0], err_r[1], err_r[2]),
transform=ax.transAxes, fontsize=10, color="w", alpha=1,
verticalalignment='top', bbox=box_propeties)
ax.text(0.03, 0.5,
'$\\bf{Attitude}$\n\t$\\theta_x = %.2f^\circ$\n\t$\\theta_y = %.2f^\circ$\n\t$\\theta_z = %.2f^\circ$'
% (euler_angles[0], euler_angles[1], euler_angles[2])
+ '\n\nError: \n\t$E_{\\theta_x} = %.3f^\circ$\n\t$E_{\\theta_y} = %.3f^\circ$'
'\n\t$E_{\\theta_z} = %.3f^\circ$'
% (err_euler[0], err_euler[1], err_euler[2]),
transform=ax.transAxes, fontsize=10, color="w", alpha=1,
verticalalignment='top', bbox=box_propeties)
# REMOVES ANY WHITESPACES FROM THE SAVED FIGURE
ax.set_axis_off()
plt.subplots_adjust(top=1, bottom=0, right=1, left=0, hspace=0, wspace=0)
plt.margins(0, 0)
ax.xaxis.set_major_locator(plt.NullLocator())
ax.yaxis.set_major_locator(plt.NullLocator())
with open("./output/param_optimization.json", "r") as f:
param_opt = f.read()
js = json.loads(param_opt)
num_LM = np.array(js["MIN_LANDMARK"])
conf_LM = np.array(js["MIN_CONFIDENCE"])
score = np.array(js["SLAB_score"])
fig = plt.figure()
ax = plt.gca()
im = ax.imshow(score, extent = [conf_LM[0] - 0.025, conf_LM[-1] + 0.025, num_LM[0] - 0.5, num_LM[-1] + 0.5],
origin = "lower", interpolation = "antialiased", aspect=0.1, vmin=np.min(score), vmax=np.max(score))
cbar = fig.colorbar(im)
cbar.ax.set_title('$\\rm{e_{SLAB}}$') # Median Normalized Pose Error
plt.xlabel('Landmark threshold confidence')
plt.ylabel('Minimum # LM')
# Best (minimum) score
ij_min = np.where(score == np.min(score))
conf_opt = conf_LM[ij_min[1][0]]
numLM_opt = num_LM[ij_min[0][0]]
print("Optimum conf score: ", conf_opt)
plt.plot(conf_opt, numLM_opt,
marker='D', markeredgecolor='black', fillstyle='none', markeredgewidth=1.5)
box_propeties = dict(boxstyle='round', facecolor='none', alpha=0.8,
edgecolor='b', linestyle='-', linewidth=1.5)
fig.savefig(os.path.join('output', 'landmark_rejection_optimization.pdf'), bbox_inches='tight')
print("Optimum min LM: ", numLM_opt)
result_file = "output/inf_results.json"
with open(result_file, "r") as f:
data = f.read()
js_pred = json.loads(data)
with open("test/test.json") as f:
data = f.read()
js_GT = json.loads(data)
with open("output/time_results_16.json") as f:
data = f.read()
time = json.loads(data)
r_pred = np.zeros((len(js_pred), 3))
q_pred = np.zeros((len(js_pred), 4))
t_det = np.zeros((len(js_pred), 1))
t_key = np.zeros((len(js_pred), 1))
t_epnp = np.zeros((len(js_pred), 1))
r_GT = np.zeros((len(js_pred), 3))
q_GT = np.zeros((len(js_pred), 4))
keypoints = np.zeros((11, 2, len(js_pred)))
bbox = np.zeros((len(js_pred), 4))
imgname = []
outlier = []
for i, _ in enumerate(js_pred):
imgname.append(js_GT[i]["filename"])
r_pred[i] = npa(js_pred[i]["tvec"])
q_pred[i] = npa(js_pred[i]["quat"])
t_det[i] = npa(time[i]["det_time"])
t_key[i] = npa(time[i]["key_time"])
t_epnp[i] = npa(js_pred[i]["epnp_time"])
r_GT[i] = npa(js_GT[i]["r_Vo2To_vbs_true"])
q_GT[i] = npa(js_GT[i]["q_vbs2tango"])
#outlier.append((js_pred[i]["outlier"]))
keypoints[:,:,i] = npa(js_pred[i]["keypoints"])
bbox[i] = npa(js_pred[i]["bbox"])
Et_vec, Et, et, eq = get_pose_error(r_GT, q_GT, r_pred, q_pred)
print("Mean translation error (x, y, z) [m]: ", np.mean(Et_vec, axis=0))
print("Mean translation error (norm) [m]: ", np.mean(Et))
print("Mean normalized translation error: ", np.mean(et))
print("Mean rotation error [deg]: ", np.rad2deg(np.mean(eq)))
print("SLAB score: ", np.mean(et + eq))
print("\nMean det time [s]: ", np.mean(t_det))
print("Mean land reg time [s]: ", np.mean(t_key))
print("Mean EPnP time: ", np.mean(t_epnp))
print("Mean total time: ", np.mean(t_det + t_key + t_epnp))
# Worst and best score:
minID = np.argmin(eq + et)
maxID = np.argmax(eq + et)
print("\nBest image: ", imgname[minID], ". Score: ", (et+eq)[minID])
print("Worst image: ", imgname[maxID], ". Score: ", (et+eq)[maxID])
# Outlier detected:
#print("\nOutlier detected in images: ", [name for i,name in enumerate(imgname) if outlier[i]])
fig = plt.figure()
ax = plt.gca()
plot_wireframe(quat2dcm(q_pred[maxID]), r_pred[maxID], keypoints[:,:,maxID], bbox[maxID], ax,
imgname[maxID], Et_vec[maxID], dcm2euler(quat2dcm(q_pred[maxID])), abs(dcm2euler(quat2dcm(q_pred[maxID])) - dcm2euler(quat2dcm(q_GT[maxID]))))
fig = plt.figure()
ax = plt.gca()
plot_wireframe(quat2dcm(q_pred[minID]), r_pred[minID], keypoints[:,:,minID], bbox[minID], ax, imgname[minID],
Et_vec[minID], dcm2euler(quat2dcm(q_pred[minID])), abs(dcm2euler(quat2dcm(q_pred[minID])) - dcm2euler(quat2dcm(q_GT[minID]))))
# Inference time plot
fig = plt.figure(figsize=(13,6))
ax = fig.gca()
ax.plot(range(len(t_det)), np.concatenate((t_det, t_key, t_epnp, t_det+t_key+t_epnp), axis=1))
ax.legend(labels = ("$t_{det}$", "$t_{key}$", "$t_{EPnP}$", "$t_{tot}$"))
ax.set_xticks(np.linspace(0, 2400, 6))
ax.set_yscale("log")
plt.xlabel("Image #")
plt.ylabel("Runtime [s]")
plt.grid(which="both")
ID_sort = np.argsort(eq + et)
# Error vs GT-distance
idx_sortDist = np.argsort(np.linalg.norm(r_GT, 2, 1))
r_GT_norm_sort = np.linalg.norm(r_GT, 2, 1)[idx_sortDist]
Et_sort = Et[idx_sortDist]
eq_sort = np.rad2deg(eq[idx_sortDist])
dist_step = 80 # i.e. we compute the mean over batches of 100 images
dist_axis = np.hstack(
(r_GT_norm_sort[np.arange(0, len(r_GT_norm_sort), dist_step)], r_GT_norm_sort[-1])
)
perc_bounds = [15.87, 84.13] # i.e. 1-sigma range
def compute_error_percentiles_vs_distance(dist_axis, dist_gt_sorted, err_vec, perc_bounds):
perc1_vec = []
perc2_vec = []
mean_err = []
mean_dist = []
for k in np.arange(len(dist_axis)-1):
d1 = dist_axis[k]
d2 = dist_axis[k+1]
idx_distStep = np.logical_and(dist_gt_sorted >= d1, dist_gt_sorted <= d2)
mean_dist.append( np.mean(dist_gt_sorted[idx_distStep]) )
values_in_dist_range = err_vec[idx_distStep]
mean_err.append( np.mean(values_in_dist_range) )
perc1_vec.append( np.percentile(values_in_dist_range, perc_bounds[0]) )
perc2_vec.append( np.percentile(values_in_dist_range, perc_bounds[1]) )
return npa(mean_dist), npa(mean_err), npa(perc1_vec), npa(perc2_vec)
# ABSOLUTE ERRORS
# Translation error
prop_cycle = plt.rcParams['axes.prop_cycle']
colors = prop_cycle.by_key()['color']
mean_dist, mean_err, p1, p2, = compute_error_percentiles_vs_distance(
dist_axis=dist_axis, dist_gt_sorted=r_GT_norm_sort,
err_vec=Et_sort*100, perc_bounds=perc_bounds
)
fig, ax = plt.subplots(2, figsize=(13,6))
ax[0].fill_between(mean_dist, p1, p2, alpha=0.3, ec='none', fc=colors[0])
ax[0].plot(mean_dist,mean_err, '-o', color=colors[0], fillstyle='none', markersize=5)
ax[0].legend(['Mean error','1$\sigma$ range'][::-1], loc=2)
ax[0].set_ylabel('Translation error [cm]')
ax[0].grid(True)
ax[0].set_axisbelow(True)
# Rotation error
mean_dist, mean_err, p1, p2, = compute_error_percentiles_vs_distance(
dist_axis=dist_axis, dist_gt_sorted=r_GT_norm_sort,
err_vec=eq_sort, perc_bounds=perc_bounds
)
ax[1].fill_between(mean_dist, p1, p2, alpha=0.3, ec='none', fc=colors[1])
ax[1].plot(mean_dist, mean_err, '-o', color=colors[1], fillstyle='none', markersize=5)
ax[1].legend(['Mean error','1$\sigma$ range'][::-1], loc=2)
ax[1].set_xlabel('Mean true distance [m]')
ax[1].set_ylabel('Quaternion error [deg]')
ax[1].grid(True)
ax[1].set_axisbelow(True)
# POSE SCORES
fig = plt.figure(figsize=[13,6])
ax = fig.gca()
# SLAB score
mean_dist, mean_err, p1, p2, = compute_error_percentiles_vs_distance(
dist_axis=dist_axis, dist_gt_sorted=r_GT_norm_sort,
err_vec=(et[idx_sortDist] + eq[idx_sortDist]), perc_bounds=perc_bounds
)
ax.fill_between(mean_dist, p1, p2, alpha=0.3, ec='none', fc=colors[2])
ax.plot(mean_dist, mean_err, '-o', color=colors[2], fillstyle='none', markersize=5)
ax.set_ylabel('SLAB score [-]')
ax.set_xlabel('Mean true distance [m]')
ax.grid(True)
ax.set_axisbelow(True)
_, mean_err, _ , _ = compute_error_percentiles_vs_distance(
dist_axis=dist_axis, dist_gt_sorted=r_GT_norm_sort,
err_vec=(et[idx_sortDist]), perc_bounds=perc_bounds
)
ax.plot(mean_dist, mean_err, '-s', color=colors[0], fillstyle='none', markersize=5)
_, mean_err, _ , _ = compute_error_percentiles_vs_distance(
dist_axis=dist_axis, dist_gt_sorted=r_GT_norm_sort,
err_vec=(eq[idx_sortDist]), perc_bounds=perc_bounds
)
ax.plot(mean_dist, mean_err, '-d', color=colors[1], fillstyle='none', markersize=5)
ax.legend(['1$\sigma$ range (total error)', 'Mean total error', 'Translation error ($e_t$)',
'Rotation error ($E_q$)'], loc=2)
# ERROR VS EARTH BACKGROUND
no_EARTH_error = []
EARTH_error = []
import re
for i in range(len(imgname)):
img_idx = int(re.search(r'\d+', imgname[i]).group())
if img_idx > earthBGStart_ID:
EARTH_error.append((et+eq)[i])
else:
no_EARTH_error.append((et+eq)[i])
print("Mean Earth error: ", np.mean(EARTH_error), " STD: ", np.std(EARTH_error))
print("Mean no Earth error: ", np.mean(no_EARTH_error), " STD: ", np.std(no_EARTH_error))
labels = ['Earth Background', 'Black Background']
means = [np.mean(EARTH_error), np.mean(no_EARTH_error)] # Replace with your mean values
std_devs = [np.std(EARTH_error), np.std(no_EARTH_error)] # Replace with your standard deviation values
fig = plt.figure()
ax = plt.gca()
for i in range(len(imgname)):
ax.clear()
plot_wireframe(quat2dcm(q_pred[i]), r_pred[i], keypoints[:,:,i], bbox[i], ax, imgname[i],
Et_vec[i], dcm2euler(quat2dcm(q_pred[i])), abs(dcm2euler(quat2dcm(q_pred[i])) - dcm2euler(quat2dcm(q_GT[i]))))
fig.savefig("./output/visualization/" + imgname[i].replace("jpg", "pdf"), bbox_inches='tight', pad_inches=0)