forked from rendezvue/PyDope
-
Notifications
You must be signed in to change notification settings - Fork 0
/
detector.py
319 lines (259 loc) · 12.1 KB
/
detector.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
# Copyright (c) 2018 NVIDIA Corporation. All rights reserved.
# This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
# https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode
"""
Revision Part for training at server 33
Feb 2021
Jiwon Gim
The original network uses 8 gpus for training. Currently Pytorch doesn't support RTX 3080. I used only one gpu (RTX 2060).
That means DataParallel (Pytorch module for using several gpus to accelerate calculation) should be turned off.
The change is reflected on line 88.
"""
'''
Contains the following classes:
- ModelData - High level information encapsulation
- ObjectDetector - Greedy algorithm to build cuboids from belief maps
'''
import time
import json
import os, shutil
import sys
import traceback
from os import path
import threading
from threading import Thread
import numpy as np
import cv2
import torch
import torch.nn as nn
import torchvision.transforms as transforms
from torch.autograd import Variable
import torchvision.models as models
from scipy import ndimage
import scipy
import scipy.ndimage as ndimage
import scipy.ndimage.filters as filters
from scipy.ndimage.filters import gaussian_filter
# Import the definition of the neural network model and cuboids
from networks import *
from cuboid_pnp_solver import *
from dope_utilities import *
#global transform for image input
transform = transforms.Compose([
# transforms.Scale(IMAGE_SIZE),
# transforms.CenterCrop((imagesize,imagesize)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
])
class ModelData(object):
'''This class contains methods for loading the neural network'''
def __init__(self, name="", net_path="", gpu_id=0):
self.name = name
self.net_path = net_path # Path to trained network model
self.net = None # Trained network
self.gpu_id = gpu_id
def get_net(self):
'''Returns network'''
if not self.net:
self.load_net_model()
return self.net
def load_net_model(self):
'''Loads network model from disk'''
if not self.net and path.exists(self.net_path):
self.net = self.load_net_model_path(self.net_path)
if not path.exists(self.net_path):
print("ERROR: Unable to find model weights: '{}'".format(
self.net_path))
exit(0)
def load_net_model_path(self, path):
'''Loads network model from disk with given path'''
model_loading_start_time = time.time()
print("Loading DOPE model '{}'...".format(path))
device = torch.device("cuda:" + str(self.gpu_id))
net = DopeNetwork()
net = net.to(device) # For model not trained with dataparallel
# net = torch.nn.DataParallel(net, [0]).cuda() # Disable dataparallel
net.load_state_dict(torch.load(path))
net.eval()
print(' Model loaded in {} seconds.'.format(
time.time() - model_loading_start_time))
return net
def __str__(self):
'''Converts to string'''
return "{}: {}".format(self.name, self.net_path)
#================================ ObjectDetector ================================
class ObjectDetector(object):
'''This class contains methods for object detection'''
@staticmethod
def detect_object_in_image(net_model, pnp_solver, in_img, config, gpu_id=0):
'''Detect objects in a image using a specific trained network model'''
if in_img is None:
return []
# Run network inference
image_tensor = transform(in_img)
device = torch.device("cuda:" + str(gpu_id))
image_torch = Variable(image_tensor).to(device).unsqueeze(0)
out, seg = net_model(image_torch)
vertex2 = out[-1][0]
aff = seg[-1][0]
# Find objects from network output
detected_objects = ObjectDetector.find_object_poses(vertex2, aff, pnp_solver, config)
return detected_objects
@staticmethod
def find_object_poses(vertex2, aff, pnp_solver, config):
'''Detect objects given network output'''
# Detect objects from belief maps and affinities
objects, all_peaks = ObjectDetector.find_objects(vertex2, aff, config)
detected_objects = []
obj_name = pnp_solver.object_name
for obj in objects:
# Run PNP
points = obj[1] + [(obj[0][0]*8, obj[0][1]*8)]
cuboid2d = np.copy(points)
location, quaternion, projected_points = pnp_solver.solve_pnp(points)
# Save results
detected_objects.append({
'name': obj_name,
'location': location,
'quaternion': quaternion,
'cuboid2d': cuboid2d,
'projected_points': projected_points,
'score': obj[-1]
})
return detected_objects
@staticmethod
def find_objects(vertex2, aff, config, numvertex=8):
'''Detects objects given network belief maps and affinities, using heuristic method'''
all_peaks = []
peak_counter = 0
for j in range(vertex2.size()[0]):
belief = vertex2[j].clone()
map_ori = belief.cpu().data.numpy()
map = gaussian_filter(belief.cpu().data.numpy(), sigma=config.sigma)
p = 1
map_left = np.zeros(map.shape)
map_left[p:,:] = map[:-p,:]
map_right = np.zeros(map.shape)
map_right[:-p,:] = map[p:,:]
map_up = np.zeros(map.shape)
map_up[:,p:] = map[:,:-p]
map_down = np.zeros(map.shape)
map_down[:,:-p] = map[:,p:]
peaks_binary = np.logical_and.reduce(
(
map >= map_left,
map >= map_right,
map >= map_up,
map >= map_down,
map > config.thresh_map)
)
peaks = zip(np.nonzero(peaks_binary)[1], np.nonzero(peaks_binary)[0])
# Computing the weigthed average for localizing the peaks
peaks = list(peaks)
win = 5
ran = win // 2
peaks_avg = []
for p_value in range(len(peaks)):
p = peaks[p_value]
weights = np.zeros((win,win))
i_values = np.zeros((win,win))
j_values = np.zeros((win,win))
for i in range(-ran,ran+1):
for j in range(-ran,ran+1):
if p[1]+i < 0 \
or p[1]+i >= map_ori.shape[0] \
or p[0]+j < 0 \
or p[0]+j >= map_ori.shape[1]:
continue
i_values[j+ran, i+ran] = p[1] + i
j_values[j+ran, i+ran] = p[0] + j
weights[j+ran, i+ran] = (map_ori[p[1]+i, p[0]+j])
# if the weights are all zeros
# then add the none continuous points
OFFSET_DUE_TO_UPSAMPLING = 0.4395
try:
peaks_avg.append(
(np.average(j_values, weights=weights) + OFFSET_DUE_TO_UPSAMPLING, \
np.average(i_values, weights=weights) + OFFSET_DUE_TO_UPSAMPLING))
except:
peaks_avg.append((p[0] + OFFSET_DUE_TO_UPSAMPLING, p[1] + OFFSET_DUE_TO_UPSAMPLING))
# Note: Python3 doesn't support len for zip object
peaks_len = min(len(np.nonzero(peaks_binary)[1]), len(np.nonzero(peaks_binary)[0]))
peaks_with_score = [peaks_avg[x_] + (map_ori[peaks[x_][1],peaks[x_][0]],) for x_ in range(len(peaks))]
id = range(peak_counter, peak_counter + peaks_len)
peaks_with_score_and_id = [peaks_with_score[i] + (id[i],) for i in range(len(id))]
all_peaks.append(peaks_with_score_and_id)
peak_counter += peaks_len
objects = []
# Check object centroid and build the objects if the centroid is found
for nb_object in range(len(all_peaks[-1])):
if all_peaks[-1][nb_object][2] > config.thresh_points:
objects.append([
[all_peaks[-1][nb_object][:2][0],all_peaks[-1][nb_object][:2][1]],
[None for i in range(numvertex)],
[None for i in range(numvertex)],
all_peaks[-1][nb_object][2]
])
# Working with an output that only has belief maps
if aff is None:
if len (objects) > 0 and len(all_peaks)>0 and len(all_peaks[0])>0:
for i_points in range(8):
if len(all_peaks[i_points])>0 and all_peaks[i_points][0][2] > config.threshold:
objects[0][1][i_points] = (all_peaks[i_points][0][0], all_peaks[i_points][0][1])
else:
# For all points found
for i_lists in range(len(all_peaks[:-1])):
lists = all_peaks[i_lists]
for candidate in lists:
if candidate[2] < config.thresh_points:
continue
i_best = -1
best_dist = 10000
best_angle = 100
for i_obj in range(len(objects)):
center = [objects[i_obj][0][0], objects[i_obj][0][1]]
# integer is used to look into the affinity map,
# but the float version is used to run
point_int = [int(candidate[0]), int(candidate[1])]
point = [candidate[0], candidate[1]]
# look at the distance to the vector field.
v_aff = np.array([
aff[i_lists*2,
point_int[1],
point_int[0]].data.item(),
aff[i_lists*2+1,
point_int[1],
point_int[0]].data.item()]) * 10
# normalize the vector
xvec = v_aff[0]
yvec = v_aff[1]
norms = np.sqrt(xvec * xvec + yvec * yvec)
xvec/=norms
yvec/=norms
v_aff = np.concatenate([[xvec],[yvec]])
v_center = np.array(center) - np.array(point)
xvec = v_center[0]
yvec = v_center[1]
norms = np.sqrt(xvec * xvec + yvec * yvec)
xvec /= norms
yvec /= norms
v_center = np.concatenate([[xvec],[yvec]])
# vector affinity
dist_angle = np.linalg.norm(v_center - v_aff)
# distance between vertexes
dist_point = np.linalg.norm(np.array(point) - np.array(center))
if dist_angle < config.thresh_angle \
and best_dist > 1000 \
or dist_angle < config.thresh_angle \
and best_dist > dist_point:
i_best = i_obj
best_angle = dist_angle
best_dist = dist_point
if i_best is -1:
continue
if objects[i_best][1][i_lists] is None \
or best_angle < config.thresh_angle \
and best_dist < objects[i_best][2][i_lists][1]:
objects[i_best][1][i_lists] = ((candidate[0])*8, (candidate[1])*8)
objects[i_best][2][i_lists] = (best_angle, best_dist)
return objects, all_peaks