forked from WongKinYiu/yolov7
-
Notifications
You must be signed in to change notification settings - Fork 0
/
deployHumveeImPogging.py
391 lines (292 loc) · 14.6 KB
/
deployHumveeImPogging.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
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
#!/usr/bin/env python
import argparse
# Libraries for deploying CNN
import numpy as np
import torch
import torch.backends.cudnn as cudnn
from numpy import random
# working VM library
from VMheader import VectorFormatter, BinCompletion
from models.experimental import attempt_load
from utils.datasets import LoadStreams, LoadImages
from utils.general import check_img_size, check_imshow, non_max_suppression, set_logging, clip_coords
from utils.torch_utils import select_device
import cv2
"""
"""
# TODO try other sat nets and see if any look like humvyy results. Also try tiny humvee and see if that one works better, add more presepectives to sat to make up for mismatching bits
# function stolen from utils general
def resizeing_corrector(locations, origonal_dim, network_dim):
return locations
def scale_coords(img1_shape, pred_list, img0_shape, ratio_pad=None):
# Rescale coords (xyxy) from img1_shape to img0_shape
gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1]) # gain = old / new
pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2 # wh padding
# print(img1_shape)
# print(img0_shape)
# print(pad)
# print((gain))
pred = []
for coords in pred_list:
coords[1] = round((coords[1] - pad[0]) / gain) # x padding
coords[3] = round((coords[3] - pad[0]) / gain) # x padding
coords[2] = round((coords[2] - pad[1]) / gain) # y padding
coords[4] = round((coords[4] - pad[1]) / gain) # y padding
pred.append(coords)
return pred
def aflattenerthingy(px, py):
# x = list(zip(px,py))
x = [px, py]
x = flatten(x)
x = np.array(x)
x = x.reshape((int(len(x) / 2), 2), order='F')
return x
def acrop_2_sat_pos(positions, path_image, shift):
path_image = path_image.split('/')
path_image = path_image[-1].split('_')
y_change = int(path_image[2])
x_change = int(path_image[-1].split('.')[0])
positions = positions + [y_change * shift, x_change * shift]
return positions
def deg_to_dms(deg): # degrees, minutes, seconds. Returns string
d = int(deg)
md = abs(deg - d) * 60
m = float(md)
return str(d) + str(m)
def flatten(l):
return [item for sublist in l for item in sublist]
def hamming2(s1, s2):
"""Calculate the Hamming distance between two bit strings"""
# assert len(s1) == len(s2)
return sum(c1 != c2 for c1, c2 in zip(s1, s2))
def pred_2_list(pred):
myorder = [5, 0, 1, 2, 3]
pred = pred.tolist()
idx = 0
for vector in pred:
pred[idx] = [vector[i] for i in myorder]
idx += 1
# print(pred)
return pred
def main():
humve_source, humve_weights, neighbor, source, weights, sat_image = opt.source, opt.weights, opt.descriptor_len, opt.sat_source, opt.sat_weights, opt.sat_image
# Paths for satellite weights and images
# weights = './satwts/yolov712/weights/best.pt'
# source = './satellite/cropped_300_scres/'
# # source = './vermontsim_5_translation/cropped_600'
# Storing values across all reviewed images (probably can go) -IP
# These are used to perform the binning process across the whole thing
vector_sat_total = []
# pos_in_sat = [] # Moved to individialy for each sat crop image
cropped_true_match = []
cropped_imdist = []
cropped_imangle = []
cropped_feature_class = []
cropped_feature_closest = []
# Initialize
set_logging()
device = select_device(device='0')
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
stride = int(model.stride.max()) # model stride
imgsz = 640
imgsz = check_img_size(imgsz, s=stride) # check img_size
if half:
model.half() # to FP16
# Set Dataloader
vid_path, vid_writer = None, None
webcam = False
if webcam:
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz, stride=stride)
else:
dataset = LoadImages(source, img_size=imgsz, stride=stride)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
old_img_w = old_img_h = imgsz
# Generating dictionary of satellite image crops so that each classes information is saved for each image
satellite_obj = {}
for path, img, im0s, vid_cap in dataset:
sat_img = img
sat_imgz_size = im0s.shape
# print(path)
sat_name = 'sat_{}'.format(path)
satellite_obj[sat_name] = VectorFormatter('satellite')
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
with torch.no_grad(): # Calculating gradients would cause a GPU memory leak
pred = model(img, augment=True)[0]
# Apply NMS, selecting only the classes that we want to use.
pred = non_max_suppression(pred, .5, .3, classes=(0,2), agnostic='store_true')
pred = pred[0]
pred = pred.detach().cpu().numpy()
pred = pred_2_list(pred)
pred = scale_coords(img.shape[2:], pred, im0s.shape) # resscales image to origonal size
satellite_obj[sat_name].cnn_initialization(pred, im0s.shape)
if len(satellite_obj[sat_name].p_x) > neighbor:
satellite_obj[sat_name].k_d_tree_test(neighbor)
cropped_imdist.append(satellite_obj[sat_name].im_dist)
cropped_imangle.append(satellite_obj[sat_name].im_angle)
cropped_feature_closest.append(satellite_obj[sat_name].feature_closest)
cropped_feature_class.append(satellite_obj[sat_name].feature_class)
# Performing binning process. Creates histogram of distribution for max vector definition effectiveness and stuff
binn = BinCompletion('binn')
binn.bin_initialize(flatten(cropped_imdist), flatten(cropped_imangle), .8, .8)
# Generating the feature descriptors/vectors for each sat image
for name in satellite_obj:
satellite_obj[name].feature_vectors = []
# Check to see if the satellite image has any feature to have descriptors made for
if len(satellite_obj[name].feature_class) > neighbor:
for j in range(len(satellite_obj[name].feature_class)):
satellite_obj[name].feature_vectors.append(
binn.vector_def(j, satellite_obj[name].feature_class, satellite_obj[name].im_dist,
satellite_obj[name].im_angle, satellite_obj[name].feature_closest)[0])
vector_sat_total.append(satellite_obj[name].feature_vectors)
vector_sat_total = flatten(vector_sat_total)
# start Drone Nural network
# Paths for satellite weights and images
# weights = './HumWts/yolov7_humvee6403/weights/best.pt'
# source = './sbtestimages/' # sb tester
source = './TrialRunIPDVermont/images_from_drone/'
# humve_weights = './satwts/yolov712/weights/best.pt'
# humve_source = './vermontsim_5_translation/cropped_800'
# Initialize
set_logging()
device = select_device(device='0')
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(humve_weights, map_location=device) # load FP32 model
stride = int(model.stride.max()) # model stride
imgsz = 640 # Humvee
imgsz = check_img_size(imgsz, s=stride) # check img_size
if half:
model.half() # to FP16
# Set Dataloader
vid_path, vid_writer = None, None
webcam = False
if webcam:
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(humve_source, img_size=imgsz, stride=stride)
else:
dataset = LoadImages(humve_source, img_size=imgsz, stride=stride)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
old_img_h = old_img_w = imgsz
hammingtotal = []
drone_file_idx = 0
for path, img, im0s, vid_cap in dataset:
drone = VectorFormatter('drone')
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
with torch.no_grad(): # Calculating gradients would cause a GPU memory leak
pred = model(img, augment=True)[0]
# Apply NMS
# First Check for humvee, if not there don't even continue
humvee_pred = non_max_suppression(pred, .2, .45, classes=6, agnostic='store_true')
# print(pred.size)
humvee_pred = humvee_pred[0]
humvee_pred = humvee_pred.detach().cpu().numpy()
humvee_pred = pred_2_list(humvee_pred)
humvee_pred = scale_coords(img.shape[2:], humvee_pred, im0s.shape)
##############insert if statement for if a humvee is detected
if not len(humvee_pred):
pred = non_max_suppression(pred, .5, .3, classes=(0, 2), agnostic='store_true')
# print(pred.size)
pred = pred[0]
pred = pred.detach().cpu().numpy()
pred = pred_2_list(pred)
pred = scale_coords(img.shape[2:], pred, im0s.shape)
drone.cnn_initialization(pred, im0s.shape)
vector_drone = []
# Checks if enough features are even detected to create a correspondence, the plus 1 is for the humvee
if len(drone.p_x) > neighbor:
drone.k_d_tree_test(neighbor)
# if drone.humvee_detected: # If humvee is detected move on, otherwise go back
drone.pos_in_sat = np.copy(aflattenerthingy(drone.p_x, drone.p_y))
for i in range(len(drone.feature_class)):
vector_drone.append(
binn.vector_def(i, drone.feature_class, drone.im_dist, drone.im_angle,
drone.feature_closest)[0])
threshold = 2
# matched_idx = np.zeros(shape=(len(vector_drone), 3)) # stored indexes [[idx sat, idx drone],.....,]
matched_idx = []
for i in range(len(vector_drone)):
best_rating = threshold
matched_idx.append([])
j = 0
for name in satellite_obj:
# print(name)
k = 0
for vector in satellite_obj[name].feature_vectors:
# if vector_drone[i][0:12] == vector_sat_total[j][0:12]:
# compair drone to sat, keep only features of the same classes
rating = hamming2(vector_drone[i], vector)
if rating <= best_rating:
# print(rating,vector_drone[i],vector_sat_total[j])
best_rating = rating
matched_idx[i] = [i, name, k]
else:
pass
k += 1
j += 1
# Show matched portions in original image
# print(matched_idx)
# satimgpath = './vermontjpg2.JPG'
# satimgpath = './lebelingtime/9-1-2021_crop_testsite.png'
satimgpath = sat_image
satimg = cv2.imread(satimgpath)
# print(matched_idx)
print(path)
for i in matched_idx:
if i: # Check if list is empty IE no matches
pos_temp = aflattenerthingy(satellite_obj[i[1]].p_x, satellite_obj[i[1]].p_y)
satellite_obj[i[1]].pos_in_sat = np.copy(acrop_2_sat_pos(pos_temp, i[1], shift=60)) #60 if sb, 350 if vmt
print(i[1])
print(int(drone.pos_in_sat[i[0], 1]), int(drone.pos_in_sat[i[0], 0]))
print(int(satellite_obj[i[1]].pos_in_sat[i[2], 1]), int(satellite_obj[i[1]].pos_in_sat[i[2], 0]))
cv2.circle(im0s, (int(drone.pos_in_sat[i[0], 0]), int(drone.pos_in_sat[i[0], 1])),
radius=10, color=(i[0] * 255 / len(matched_idx), 0, 255 -(i[0] * 255 / len(matched_idx))), thickness=4)
cv2.circle(satimg,
(int(satellite_obj[i[1]].pos_in_sat[i[2], 0]),
int(satellite_obj[i[1]].pos_in_sat[i[2], 1])),
radius=10, color=(i[0] * 255 / len(matched_idx), 0, 255 -(i[0] * 255 / len(matched_idx))), thickness=-4)
cv2.imwrite(opt.results+'/drone{}.png'.format(drone_file_idx), im0s)
cv2.imwrite(opt.results+'/sat{}.png'.format(drone_file_idx), satimg)
drone_file_idx +=1
# cv2.imshow('69', im0s)
# cv2.waitKey(1) # 1 millisecond
# cv2.imshow('69', satimg)
# cv2.waitKey(1) # 1 millisecond
# im0s = cv2.resize(im0s, (satimg.shape[1], satimg.shape[0]), interpolation=cv2.INTER_AREA)
# # Verti = np.concatenate((im0s, satimg), axis=0)
# plt.figure()
# plt.subplot(121)
# plt.imshow(im0s)
# plt.subplot(122)
# plt.imshow(satimg)
# plt.show()
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--weights', nargs='+', type=str, default='yolov7.pt', help='model.pt path(s)')
parser.add_argument('--sat_weights', nargs='+', type=str, default='yolov7.pt', help='model.pt path(s)')
parser.add_argument('--source', type=str, default='inference/images', help='source') # file/folder, 0 for webcam
parser.add_argument('--sat_source', type=str, default='inference/images', help='source') # file/folder, 0 for webcam
parser.add_argument('--sat_image', type=str, default='inference/images', help='source') # file/folder, 0 for webcam
parser.add_argument('--results', type=str, default='./results', help='results folder') # file/folder, 0 for webcam
parser.add_argument('--descriptor_len', type=int, default=7, help='how many features min detected and length of descriptor')
opt = parser.parse_args()
with torch.no_grad():
main()