def test():
os.system('rm -rf tmp/pos/*')
os.system('rm -rf tmp/neg/*')
os.system('rm -rf tmp/part/*')
logger.info('Load WIDER')
train_data, val_data = load_wider()
print("TRAIN DATA: ", len(train_data))
img_path, bboxes = train_data[np.random.choice(len(train_data))]
bboxes = np.asarray(bboxes)
#pdb.set_trace()
img = cv2.imread(img_path, cv2.IMREAD_COLOR)
detector = JfdaDetector(cfg.PROPOSAL_NETS['r'])
#print("HERE??>> 529")
negatives, positives, part = proposal(img, bboxes, detector)
logger.info('%d gt_bboxes', len(bboxes))
logger.info('%d negatives, %d positives, %d part', len(negatives),
len(positives), len(part))
for i, (data[1], bbox_target) in enumerate(positives):
cv2.imwrite(longPath + 'tmp/pos/%03d.png' % i, data)
for i, (data) in enumerate(negatives):
cv2.imwrite(longPath + 'tmp/neg/%03d.png' % i, data[0])
#pdb.set_trace()
for i, (data[1], bbox_target) in enumerate(part):
cv2.imwrite('tmp/part/%03d.png' % i, data[0])
cv2.imwrite('tmp/test.png', img)
def get_detector():
nets = cfg.PROPOSAL_NETS[cfg.NET_TYPE]
if nets is None or not cfg.USE_DETECT:
detector = None
else:
if cfg.GPU_ID >= 0:
caffe.set_mode_gpu()
caffe.set_device(cfg.GPU_ID)
else:
caffe.set_mode_cpu()
detector = JfdaDetector(nets)
return detector
def main(args):
net = [
'proto/p.prototxt',
'model/p.caffemodel',
'proto/r.prototxt',
'model/r.caffemodel',
'proto/o.prototxt',
'model/o.caffemodel',
'proto/l.prototxt',
'model/l.caffemodel',
]
if args.minicaffe:
from jfda.mdetector import MiniCaffeDetector
detector = MiniCaffeDetector(net)
else:
detector = JfdaDetector(net)
if args.pnet_single:
detector.set_pnet_single_forward(True)
param = {
'ths': [0.6, 0.7, 0.8],
'factor': 0.709,
'min_size': 24,
}
timer = Timer()
def gen(img, bboxes, outname):
for i in range(len(bboxes)):
x1, y1, x2, y2, score = bboxes[i, :5]
x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)
cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255), 2)
cv2.putText(img, '%.03f' % score, (x1, y1), cv2.FONT_HERSHEY_PLAIN,
1, (0, 255, 0))
# landmark
landmark = bboxes[i, 9:].reshape((5, 2))
for j in range(5):
x, y = landmark[j]
x, y = int(x), int(y)
cv2.circle(img, (x, y), 2, (0, 255, 0), -1)
cv2.imwrite(outname, img)
with open('demo.txt', 'r') as fin:
for line in fin.readlines():
fp = line.strip()
dn = os.path.dirname(fp)
fn = os.path.basename(fp).split('.')[0]
img = cv2.imread(fp, cv2.IMREAD_COLOR)
timer.tic()
bb, ts = detector.detect(img, debug=True, **param)
timer.toc()
print 'detect %s costs %.04lfs' % (fp, timer.elapsed())
print 'image size = (%d x %d), s1: %.04lfs, s2: %.04lfs, s3: %.04lfs, s4: %.04lf' % (
img.shape[0], img.shape[1], ts[0], ts[1], ts[2], ts[3])
print 'bboxes, s1: %d, s2: %d, s3: %d, s4: %d' % (len(
bb[0]), len(bb[1]), len(bb[2]), len(bb[3]))
out1 = '%s/%s_stage1.jpg' % (dn, fn)
out2 = '%s/%s_stage2.jpg' % (dn, fn)
out3 = '%s/%s_stage3.jpg' % (dn, fn)
out4 = '%s/%s_stage4.jpg' % (dn, fn)
gen(img.copy(), bb[0], out1)
gen(img.copy(), bb[1], out2)
gen(img.copy(), bb[2], out3)
gen(img.copy(), bb[3], out4)
def main(args):
if args.net == 'p':
net = ['proto/p.prototxt', 'model/p.caffemodel']
elif args.net == 'r':
net = [
'proto/p.prototxt', 'model/p.caffemodel', 'proto/r.prototxt',
'model/r.caffemodel'
]
else:
net = [
'proto/p.prototxt',
'model/p.caffemodel',
'proto/r.prototxt',
'model/r.caffemodel',
'proto/o.prototxt',
'model/o.caffemodel',
]
detector = JfdaDetector(net)
if args.pnet_single:
detector.set_pnet_single_forward(True)
logger = get_logger()
counter = 0
timer = Timer()
param = {
'ths': [0.5, 0.5, 0.4],
'factor': 0.709,
'min_size': 24,
}
total_time = 0.
total_width = 0
total_height = 0
for i in range(10):
logger.info('Process FOLD-%02d', i)
txt_in = cfg.FDDB_DIR + '/FDDB-folds/FDDB-fold-%02d.txt' % (i + 1)
txt_out = cfg.FDDB_DIR + '/result/fold-%02d-out.txt' % (i + 1)
answer_in = cfg.FDDB_DIR + '/FDDB-folds/FDDB-fold-%02d-ellipseList.txt' % (
i + 1)
fin = open(txt_in, 'r')
fout = open(txt_out, 'w')
ain = open(answer_in, 'r')
for line in fin.readlines():
line = line.strip()
in_file = cfg.FDDB_DIR + '/images/' + line + '.jpg'
out_file = cfg.FDDB_DIR + '/result/images/' + line.replace(
'/', '-') + '.jpg'
counter += 1
img = cv2.imread(in_file, cv2.IMREAD_COLOR)
timer.tic()
bboxes = detector.detect(img, **param)
timer.toc()
total_time += timer.elapsed()
h, w = img.shape[:-1]
total_height += h
total_width += w
logger.info(
'Detect %04d th image costs %.04lfs, average %.04lfs, %d x %d',
counter, timer.elapsed(), total_time / counter,
total_height / counter, total_width / counter)
fout.write('%s\n' % line)
fout.write('%d\n' % len(bboxes))
for bbox in bboxes:
x1, y1, x2, y2, score = bbox[:5]
fout.write('%lf %lf %lf %lf %lf\n' %
(x1, y1, x2 - x1, y2 - y1, score))
# draw ground truth
ain.readline() # remove fname
n = int(ain.readline().strip())
for i in range(n):
line = ain.readline().strip()
components = [float(_) for _ in line.split(' ')[:5]]
major_axis_radius, minor_axis_radius, angle, center_x, center_y = components
angle = angle / 3.1415926 * 180.
center_x, center_y = int(center_x), int(center_y)
major_axis_radius, minor_axis_radius = int(
major_axis_radius), int(minor_axis_radius)
cv2.ellipse(img, (center_x, center_y),
(major_axis_radius, minor_axis_radius), angle, 0,
360, (255, 0, 0), 2)
# draw and save
for bbox in bboxes:
x1, y1, x2, y2, score = bbox[:5]
x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)
cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255), 2)
cv2.putText(img, '%.03f' % score, (x1, y1),
cv2.FONT_HERSHEY_PLAIN, 1, (0, 255, 0))
landmark = bbox[9:].reshape((5, 2))
for x, y in landmark:
x, y = int(x), int(y)
cv2.circle(img, (x, y), 2, (0, 255, 0), -1)
cv2.imwrite(out_file, img)
fin.close()
fout.close()
ain.close()