-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathvalidate_approach_synthhands.py
338 lines (290 loc) · 15.1 KB
/
validate_approach_synthhands.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
# exampel calls
# -i female_object/seq01/cam01/01/00000000 -r /home/paulo/SynthHands_Release/ --halnet /home/paulo/muellericcv2017/trainednets/trained_HALNet_1493752625_for_valid_38000.pth.tar --jornet /home/paulo/muellericcv2017/trainednets/trained_JORNet_1662451312_for_valid_70000.pth.tar
# -i Fruits/color_on_depth/image_00000 -r /home/paulo/EgoDexter/data/ --halnet /home/paulo/muellericcv2017/trainednets/trained_HALNet_1493752625_.pth.tar --jornet /home/paulo/muellericcv2017/trainednets/trained_JORNet_1662451312_for_valid_30000.pth.tar
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
import numpy as np
import io_image
import trainer
import synthhands_handler
import argparse
import converter as conv
import HALNet, JORNet
import time
import camera
import visualize
MAX_NUM_EXAMPLES = 30
IMG_RES = (320, 240)
def print_time(str_, time_diff):
print(str_ + str(round(time_diff*1000)) + ' ms')
def plot_joints(joints_colorspace, show_legend=True, linewidth=4):
num_joints = joints_colorspace.shape[0]
joints_colorspace = conv.numpy_swap_cols(joints_colorspace, 0, 1)
plt.plot(joints_colorspace[0, 1], joints_colorspace[0, 0], 'ro', color='C0')
plt.plot(joints_colorspace[0:2, 1], joints_colorspace[0:2, 0], 'ro-', color='C0', linewidth=linewidth)
joints_names = ['Thumb', 'Index', 'Middle', 'Ring', 'Little']
legends = []
if show_legend:
palm_leg = mpatches.Patch(color='C0', label='Palm')
legends.append(palm_leg)
for i in range(4):
plt.plot([joints_colorspace[0, 1], joints_colorspace[(i * 4) + 5, 1]],
[joints_colorspace[0, 0], joints_colorspace[(i * 4) + 5, 0]], 'ro-', color='C0', linewidth=linewidth)
for i in range(num_joints - 1):
if (i + 1) % 4 == 0:
continue
color = 'C' + str(int(np.ceil((i + 1) / 4)))
plt.plot(joints_colorspace[i + 1:i + 3, 1], joints_colorspace[i + 1:i + 3, 0], 'ro-', color=color, linewidth=linewidth)
if show_legend and i % 4 == 0:
joint_name = joints_names[int(np.floor((i+1)/4))]
legends.append(mpatches.Patch(color=color, label=joint_name))
if show_legend:
plt.legend(handles=legends)
return joints_colorspace
def get_image_name(image_basename, ix, dataset_name):
str_to_add = str(ix)
if ix == 0:
ix = 1
algs = int(np.log10(ix))
image_basename = image_basename[:-(algs+1)] + str_to_add
return image_basename
def get_image_as_data(dataset_folder, input_img_namebase, dataset_name, img_res):
data = []
if dataset_name == 'SynthHands_Release':
data = synthhands_handler._get_data(dataset_folder, input_img_namebase, img_res)
elif dataset_name == 'EgoDexter':
data = egodexter_handler.get_data(dataset_folder, input_img_namebase, img_res=img_res)
return data
def load_examples_to_memory(num_labels, dataset):
data = []
img_paths = []
for i in range(num_labels):
img_paths.append(dataset.filenamebases[i])
img_data, img_labels = dataset[i]
labels_colorspace, labels_jointvec, _, handroot_groundtruth = img_labels
data.append((img_data, (labels_colorspace, labels_jointvec, handroot_groundtruth)))
return data, img_paths
def parse_args():
parser = argparse.ArgumentParser(description='Train a hand-tracking deep neural network')
parser.add_argument('-r', dest='dataset_folder', default='', type=str, required=True,
help='Dataset folder')
parser.add_argument('--halnet', dest='halnet_filepath', type=str, required=True,
help='Filepath to trained HALNet checkpoint')
parser.add_argument('--jornet', dest='jornet_filepath', type=str, required=True,
help='Filepath to trained HALNet checkpoint')
parser.add_argument('--cuda', dest='use_cuda', action='store_true', default=False,
help='Whether to use cuda for training')
parser.add_argument('-o', dest='output_filepath', default='',
help='Output file for logging')
return parser.parse_args()
def print_divisor(num=100):
print('-' * num)
args = parse_args()
dataset_name = args.dataset_folder.split('/')[-2]
synthhands = synthhands_handler.SynthHandsDataset(root_folder=args.dataset_folder, type_='full', heatmap_res=IMG_RES)
num_examples = min(MAX_NUM_EXAMPLES, len(synthhands))
print_divisor()
print('Arguments')
print('\tDataset root: {}'.format(args.dataset_folder))
print('\tHALNet filepath: {}'.format(args.halnet_filepath))
print('\tJORNet filepath: {}'.format(args.jornet_filepath))
print('\tUsing CUDA: {}'.format(args.use_cuda))
print_divisor()
print('Dataset: ')
print('\tSize of dataset: {}'.format(len(synthhands)))
print('\tNumber of examples to process: {}'.format(num_examples))
start = time.time()
dataset_data, img_paths = load_examples_to_memory(num_examples, synthhands)
print_time('\tLoaded examples to memory: ', time.time() - start)
print_divisor()
print('Neural networks: ')
# load nets
start = time.time()
halnet, _, _, _ = trainer.load_checkpoint(filename=args.halnet_filepath,
model_class=HALNet.HALNet,
use_cuda=args.use_cuda)
print_time('\tHALNet loaded: ', time.time() - start)
start = time.time()
jornet, _, _, _ = trainer.load_checkpoint(filename=args.jornet_filepath,
model_class=JORNet.JORNet,
use_cuda=args.use_cuda)
print_time('\tJORNet loaded: ', time.time() - start)
print_divisor()
losses_halnet_per_joints = np.zeros((num_examples, 21))
losses_halnet_joints_tot = []
losses_jornet_per_joints = np.zeros((num_examples, 21))
losses_jornet_joints_tot = []
losses_jornet_per_joints_depth = np.zeros((num_examples, 21))
losses_jornet_depth_tot = []
num_valid_loss_iter = 0
tot_loss = 0
for example_ix in range(num_examples):
start_beg = time.time()
print('Processing example #{}: {}'.format(example_ix, img_paths[example_ix]))
dataset_datum = dataset_data[example_ix]
img_data, img_labels = dataset_datum
img_numpy = img_data.data.numpy()
img_labels_2D, img_labels_3D, handroot_groundtruth = img_labels
img_labels_3D = img_labels_3D.data.numpy().reshape((21, 3))
print('\tHALNet:')
start = time.time()
halnet_input = conv.data_to_batch(img_data)
print_time('\t\tHALNet image convertion: ', time.time() - start)
start = time.time()
output_halnet = halnet(halnet_input)
print_time('\t\tHALNet pass: ', time.time() - start)
halnet_main_out = output_halnet[3][0].data.numpy()
halnet_handroot = np.array(np.unravel_index(np.argmax(halnet_main_out[0]), IMG_RES))
print('\t\t\tHALNet hand root:\t{}'.format(halnet_handroot))
# get halnet joints in colorspace from heatmaps
halnet_joints_colorspace = conv.heatmaps_to_joints_colorspace(halnet_main_out)
print('\tHALNet joint pixel loss:')
num_valid_loss = 0
loss_halnet_joints = 0
for i in range(21):
curr_loss = np.linalg.norm(img_labels_2D[i, :] - halnet_joints_colorspace[i, :])
loss_halnet_joints += curr_loss
losses_halnet_per_joints[example_ix, i] = curr_loss
print('\t\t\tJoint {}: {}\t{} : {}'.format(i, img_labels_2D[i, :], halnet_joints_colorspace[i, :], curr_loss))
losses_halnet_joints_tot.append(loss_halnet_joints / 21)
print('\t\tAverage loss for HALNet joints: {}'.format(losses_halnet_joints_tot[-1]))
print_divisor()
#fig = visualize.plot_image(img_numpy)
#visualize.plot_joints(halnet_joints_colorspace, fig=fig)
#visualize.show()
print('\tJORNet:')
start = time.time()
data_crop, crop_coords, _, _ = io_image.crop_image_get_labels(img_numpy, halnet_joints_colorspace)
batch_jornet = conv.data_to_batch(data_crop)
print_time('\t\tJORNet image conversion: ', time.time() - start)
start = time.time()
output_jornet = jornet(batch_jornet)
print_time('\t\tJORNet pass: ', time.time() - start)
_, img_labels_2D_cropped = io_image.get_labels_cropped_heatmaps(
img_labels_2D, joint_ixs=range(21), crop_coords=crop_coords, heatmap_res=(128, 128))
print('\tImage labels (2D cropped to (128, 128)):')
for i in range(21):
depth_const = img_numpy[3, int(img_labels_2D_cropped[i, 0]), int(img_labels_2D_cropped[i, 1])]
print('\t\tFinger tip {}: {}'.format(i, img_labels_2D_cropped[i, :]))
print_divisor()
print('\tJORNet joint pixel loss:')
output_jornet_heatmaps_main = output_jornet[3][0].data.numpy()
jornet_joints_colorspace = conv.heatmaps_to_joints_colorspace(output_jornet_heatmaps_main)
num_valid_loss = 0
loss_jornet_joints = 0
for i in range(21):
curr_loss = np.linalg.norm(img_labels_2D_cropped[i, :] - jornet_joints_colorspace[i, :])
loss_jornet_joints += curr_loss
losses_jornet_per_joints[example_ix, i] = curr_loss
print('\t\t\tJoint {}: {}\t{} : {}'.format(i, img_labels_2D_cropped[i, :], jornet_joints_colorspace[i, :], curr_loss))
losses_jornet_joints_tot.append(loss_jornet_joints / 21)
print('\t\tAverage loss for HALNet joints: {}'.format(losses_jornet_joints_tot[-1]))
print_divisor()
#fig = visualize.plot_image(data_crop)
#visualize.plot_joints(img_labels_2D_cropped, fig=fig)
#visualize.plot_joints(jornet_joints_colorspace, fig=fig)
#visualize.show()
output_jornet_joints_main = output_jornet[7][0].data.cpu().numpy().reshape((20, 3))
#handroot = camera.joint_color2depth(halnet_joints_colorspace[0, 0],
# halnet_joints_colorspace[0, 1],
# 200,
# synthhands_handler.DEPTH_INTR_MTX_INV)
handroot = handroot_groundtruth
jornet_joints_global = conv.jornet_local_to_global_joints(output_jornet_joints_main, handroot)
#fig, ax = visualize.plot_3D_joints(jornet_joints_global)
#visualize.plot_3D_joints(img_labels_3D, fig=fig, ax=ax)
#visualize.show()
print('\tJORNet joint depth loss:')
num_valid_loss = 0
loss_jornet_depth = 0
for i in range(21):
curr_loss = np.linalg.norm(img_labels_3D[i, :] - jornet_joints_global[i, :])
losses_jornet_per_joints_depth[example_ix, i] = curr_loss
loss_jornet_depth += curr_loss
print('\t\t\tFinger tip {}: {}\t{} : {}'.format(i, img_labels_3D[i, :],
jornet_joints_global[i, :],
curr_loss))
loss_jornet_depth /= 21
losses_jornet_depth_tot.append(loss_jornet_depth)
print('\t\tAverage loss for JORNet depth: {}'.format(loss_jornet_depth))
print_divisor()
print('\tLosses:')
print('\tHALNet')
print('\t\tAverage joint loss: {}'.format(np.mean(losses_halnet_joints_tot)))
print('\t\tStddev joint loss: {}'.format(np.std(losses_halnet_joints_tot)))
print('\t\tPer joint loss (average ; std):')
halnet_means_per_joint = [0.] * 22
halnet_err_per_joint = [0.] * 22
for joint_ix in range(21):
halnet_means_per_joint[joint_ix] = float(np.mean(losses_halnet_per_joints[:, joint_ix]))
halnet_err_per_joint[joint_ix] = float(np.std(losses_halnet_per_joints[:, joint_ix])) / 2.
print('\t\t\tJoint\t{}:\t{}\t;\t{}'.
format(joint_ix,
halnet_means_per_joint[joint_ix],
np.std(losses_halnet_per_joints[:, joint_ix])))
halnet_means_per_joint[-1] = np.mean(losses_halnet_joints_tot)
halnet_err_per_joint[-1] = np.std(losses_halnet_joints_tot)
print('\tJORNet')
print('\t\tPer joint loss (average ; std):')
jornet_means_per_joint = [0.] * 22
jornet_err_per_joint = [0.] * 22
for joint_ix in range(21):
jornet_means_per_joint[joint_ix] = float(np.mean(losses_jornet_per_joints[:, joint_ix]))
jornet_err_per_joint[joint_ix] = float(np.std(losses_jornet_per_joints[:, joint_ix])) / 2.
print('\t\t\tJoint\t{}:\t{}\t;\t{}'.
format(joint_ix,
jornet_means_per_joint[joint_ix],
np.std(losses_jornet_per_joints[:, joint_ix])))
jornet_means_per_joint[-1] = np.mean(losses_jornet_joints_tot)
jornet_err_per_joint[-1] = np.std(losses_jornet_joints_tot)
print('\t\tPer joint loss depth (average ; std):')
jornet_means_per_joint_depth = [0.] * 22
jornet_err_per_joint_depth = [0.] * 22
for joint_ix in range(21):
jornet_means_per_joint_depth[joint_ix] = float(np.mean(losses_jornet_per_joints_depth[:, joint_ix]))
jornet_err_per_joint_depth[joint_ix] = float(np.std(losses_jornet_per_joints_depth[:, joint_ix])) / 2.
print('\t\t\tJoint\t{}:\t{}\t;\t{}'.
format(joint_ix,
jornet_means_per_joint_depth[joint_ix],
np.std(losses_jornet_per_joints_depth[:, joint_ix])))
jornet_means_per_joint_depth[-1] = np.mean(losses_jornet_depth_tot)
jornet_err_per_joint_depth[-1] = np.std(losses_jornet_depth_tot)
print('\t\tAverage depth loss: {}'.format(np.mean(losses_jornet_depth_tot)))
print('\t\tStddev depth loss: {}'.format(np.std(losses_jornet_depth_tot)))
print_divisor()
num_ranges = 30
max_range = 30
frames_per_mm_range = np.zeros((num_ranges, 1))
losses_jornet_joints_tot_array = np.array(losses_jornet_joints_tot)
for i in range(num_ranges):
ixs = np.array(losses_jornet_joints_tot_array < i * (max_range/num_ranges))
frames_per_mm_range[i] = (np.sum(ixs) / len(losses_jornet_joints_tot)) * 100
print(frames_per_mm_range[i])
print_divisor()
visualize.plot_line(frames_per_mm_range, xlabel='Error threshold (pixels)', ylabel='Percentage of frames',
fontsize=30, tickwidth=10, linewidth=10)
visualize.show()
num_mm_ranges = 60
max_range = 60
frames_per_mm_range = np.zeros((num_mm_ranges, 1))
losses_jornet_depth_tot_array = np.array(losses_jornet_depth_tot)
for i in range(num_mm_ranges):
ixs = np.array(losses_jornet_depth_tot_array < i * (max_range/num_mm_ranges))
frames_per_mm_range[i] = (np.sum(ixs) / len(losses_jornet_depth_tot)) * 100
print(frames_per_mm_range[i])
print_divisor()
visualize.plot_line(frames_per_mm_range, xlabel='Error threshold (mm)', ylabel='Percentage of frames',
fontsize=30, tickwidth=10, linewidth=10)
visualize.show()
visualize.plot_per_joint_bar_chart(halnet_means_per_joint, halnet_err_per_joint, added_avg_value=True,
horizontal=True, xlabel='Joint dist loss (pixels)',
ylabel='Joint name', title='SynthHands : HALNet: Loss per Joint (pixels)')
visualize.show()
visualize.plot_per_joint_bar_chart(jornet_means_per_joint, jornet_err_per_joint, added_avg_value=True,
horizontal=True, xlabel='Joint dist loss (pixels)',
ylabel='Joint name', title='SynthHands : JORNet: Loss per Joint (pixels)')
visualize.show()
visualize.plot_per_joint_bar_chart(jornet_means_per_joint_depth, jornet_err_per_joint_depth, added_avg_value=True,
horizontal=True, xlabel='Joint dist loss (mm)',
ylabel='Joint name', title='SynthHands : JORNet: Loss per Joint (depth)')
visualize.show()