def do_fr_after_YOLO(imgs_paths):
output_dir = 'output_txt/fr_after_YOLO_kr'
# using yolo v2 - voc
YOLO_weights = 'data/vehicle-detector/yolov3.weights'
YOLO_netcfg = 'data/vehicle-detector/yolov3.cfg'
YOLO_data = 'data/vehicle-detector/coco.data'
print 'YOLOv3 weights pre-loading...'
YOLO_net = dn.load_net(YOLO_netcfg, YOLO_weights, 0)
YOLO_meta = dn.load_meta(YOLO_data)
threshold = 0.5
for img_path in imgs_paths:
print 'detecting cars in', img_path
img = cv2.imread(img_path)
results, wh = dn.detect(YOLO_net, YOLO_meta, img, threshold)
txt_file = open(
join(output_dir,
basename(splitext(img_path)[0]) + '.txt'), 'w')
if len(results) == 0:
txt_file.write('')
continue
for result in results:
if result[0] in ['car', 'bus']:
WH = np.array(img.shape[1::-1], dtype=float)
cx, cy, w, h = (np.array(result[2]) / np.concatenate(
(WH, WH))).tolist()
tl = np.array([cx - w / 2., cy - h / 2.])
br = np.array([cx + w / 2., cy + h / 2.])
label_sub = Label(tl=tl, br=br)
sub_img = crop_region(img, label_sub)
# sub_image FRD, only process the highest prob one
print '\tFRD processing...'
frd, _ = dn.detect(FR_net, FR_meta, sub_img, threshold)
if len(frd) == 0:
continue
WH_sub = np.array(sub_img.shape[1::-1], dtype=float)
cx, cy, w, h = (np.array(frd[0][2]) / np.concatenate(
(WH_sub, WH_sub))).tolist()
tl = np.array([cx - w / 2., cy - h / 2.])
br = np.array([cx + w / 2., cy + h / 2.])
label_FR = Label(tl=tl, br=br)
label_scale_up = Label(
tl=label_sub.tl() * WH + label_FR.tl() * WH_sub,
br=label_sub.tl() * WH + label_FR.br() * WH_sub)
tl = label_scale_up.tl().astype(int)
br = label_scale_up.br().astype(int)
txt_file.write(frd[0][0] + ' ' + str('%.2f' % frd[0][1]) +
' ' + str(tl[0]) + ' ' + str(tl[1]) + ' ' +
str(br[0]) + ' ' + str(br[1]) + '\n')
print '\twrote result to', join(
output_dir,
basename(splitext(img_path)[0]) + '.txt')
txt_file.close()
def augment_sample(I,pts,dim):
maxsum,maxangle = 120,np.array([80.,80.,45.])
angles = np.random.rand(3)*maxangle
if angles.sum() > maxsum:
angles = (angles/angles.sum())*(maxangle/maxangle.sum())
I = im2single(I)
iwh = getWH(I.shape)
whratio = random.uniform(2.,4.)
wsiz = random.uniform(dim*.2,dim*1.)
hsiz = wsiz/whratio
dx = random.uniform(0.,dim - wsiz)
dy = random.uniform(0.,dim - hsiz)
pph = getRectPts(dx,dy,dx+wsiz,dy+hsiz)
pts = pts*iwh.reshape((2,1))
T = find_T_matrix(pts2ptsh(pts),pph)
H = perspective_transform((dim,dim),angles=angles)
H = np.matmul(H,T)
Iroi,pts = project(I,H,pts,dim)
hsv_mod = np.random.rand(3).astype('float32')
hsv_mod = (hsv_mod - .5)*.3
hsv_mod[0] *= 360
Iroi = hsv_transform(Iroi,hsv_mod)
Iroi = np.clip(Iroi,0.,1.)
pts = np.array(pts)
if random.random() > .5:
Iroi,pts = flip_image_and_pts(Iroi,pts)
tl,br = pts.min(1),pts.max(1)
llp = Label(0,tl,br)
return Iroi,llp,pts
def fr_detect(img): print '\t\t\tdetecting front and rear using FRD..., Model:', FR_netcfg results, wh = dn.detect(FR_net, FR_meta, img, threshold) # the results will be list according to its probability , high prob -> low prob if len(results): print '\t\t\tFR detection completed' FRs = [] category = [] for i, result in enumerate(results): WH = np.array(img.shape[1::-1], dtype=float) cx, cy, w, h = (np.array(result[2]) / np.concatenate((WH, WH))).tolist() tl = np.array([cx - w / 2., cy - h / 2.]) br = np.array([cx + w / 2., cy + h / 2.]) print '\t\t\tFR number', i, 'position:', tl, br, 'prob:', result[1] FRs.append(Label(tl=tl, br=br)) category.append(result[0]) return np.array(FRs), category else: print '\t\t\tFR detection failed'
def do_fr(imgs_paths):
modelmame = basename(splitext(FR_netcfg)[0])
weight = basename(splitext(FR_weights)[0])
model_dir = join('output_txt/', modelmame + '_' + valid_dataset)
final_dir = join(model_dir, weight)
if not isdir(model_dir):
mkdir(model_dir)
if not isdir(final_dir):
mkdir(final_dir)
for img_path in imgs_paths:
print 'FRD processing on', img_path
img = cv2.imread(img_path)
frds, wh = dn.detect(FR_net, FR_meta, img, threshold)
txt_file = open(
join(final_dir,
basename(splitext(img_path)[0]) + '.txt'), 'w')
if len(frds) == 0:
txt_file.write('')
continue
for frd in frds:
cx, cy, w, h = (np.array(frd[2])).tolist()
tl = np.array([cx - w / 2., cy - h / 2.])
br = np.array([cx + w / 2., cy + h / 2.])
label_FR = Label(tl=tl, br=br)
tl = label_FR.tl().astype(int)
br = label_FR.br().astype(int)
txt_file.write(frd[0] + ' ' + str('%.2f' % frd[1]) + ' ' +
str(tl[0]) + ' ' + str(tl[1]) + ' ' + str(br[0]) +
' ' + str(br[1]) + '\n')
print '\twrote result to', join(
final_dir,
basename(splitext(img_path)[0]) + '.txt')
txt_file.close()
def getSquare(self): tl,br = self.pts.min(1),self.pts.max(1) return Label(-1,tl,br)
# R = [r for r in R if r[0] in ['front', 'rear']]
print '\t\t%d cars found' % len(R)
if len(R):
Iorig = cv2.imread(img_path)
WH = np.array(Iorig.shape[1::-1],dtype=float)
Lcars = []
# each detected vehicle
for i,r in enumerate(R):
# make the bounding box coordinate be related to the original image size
cx,cy,w,h = (np.array(r[2])/np.concatenate( (WH,WH) )).tolist()
# tl-> top left, br-> bottom right
tl = np.array([cx - w/2., cy - h/2.])
br = np.array([cx + w/2., cy + h/2.])
label = Label(0,tl,br)
Icar = crop_region(Iorig,label)
Lcars.append(label)
cv2.imwrite('%s/%s_%dcar.png' % (output_dir,bname,i),Icar)
lwrite('%s/%s_cars.txt' % (output_dir,bname),Lcars)
except:
traceback.print_exc()
sys.exit(1)
sys.exit(0)
def union_area(bb_1, bb_2):
final_tl = np.minimum.reduce([bb_1[0], bb_1[1], bb_2[0], bb_2[1]])
final_br = np.maximum.reduce([bb_1[0], bb_1[1], bb_2[0], bb_2[1]])
return Label(tl=final_tl, br=final_br)
R = [r for r in R if r[0] in ['car', 'bus']]
print '\t\t%d cars found' % len(R)
if len(R):
WH = np.array(img.shape[1::-1], dtype=float)
Lcars = []
for i, r in enumerate(R):
cx, cy, w, h = (np.array(r[2]) / np.concatenate(
(WH, WH))).tolist()
tl = np.array([cx - w / 2., cy - h / 2.])
br = np.array([cx + w / 2., cy + h / 2.])
label = Label(0, tl, br)
Lcars.append(label)
lwrite('%s/%s_cars.txt' % (output_dir, bname), Lcars)
# license plate detection
try:
# colors are BGR in opencv
YELLOW = (0, 255, 255)
RED = (0, 0, 255)
PINK = (232, 28, 232)
input_dir = output_dir
lp_threshold = 0.5
