def setUp(self):
self.link = Multibox(self.n_class, self.aspect_ratios)
xs = list()
n_bbox = 0
for ar in self.aspect_ratios:
C, H, W = np.random.randint(1, 10, size=3)
xs.append(
np.random.uniform(size=(self.batchsize, C, H,
W)).astype(np.float32))
n_bbox += H * W * (len(ar) + 1) * 2
self.xs = xs
self.n_bbox = n_bbox
def __init__(self, n_fg_class):
super(DummySSD,
self).__init__(extractor=DummyExtractor(),
multibox=Multibox(n_class=n_fg_class + 1,
aspect_ratios=((2, ), (2, 3),
(2, ))),
steps=(0.1, 0.25, 1),
sizes=(0.1, 0.25, 1, 1.2),
mean=np.array((0, 1, 2)).reshape((-1, 1, 1)))
class TestMultibox(unittest.TestCase):
def setUp(self):
self.link = Multibox(self.n_class, self.aspect_ratios)
xs = list()
n_bbox = 0
for ar in self.aspect_ratios:
C, H, W = np.random.randint(1, 10, size=3)
xs.append(
np.random.uniform(size=(self.batchsize, C, H,
W)).astype(np.float32))
n_bbox += H * W * (len(ar) + 1) * 2
self.xs = xs
self.n_bbox = n_bbox
def _check_forward(self, xs):
mb_locs, mb_confs = self.link(xs)
self.assertIsInstance(mb_locs, chainer.Variable)
self.assertIsInstance(mb_locs.data, type(xs[0]))
self.assertEqual(mb_locs.shape, (self.batchsize, self.n_bbox, 4))
self.assertEqual(mb_locs.dtype, xs[0].dtype)
self.assertIsInstance(mb_confs, chainer.Variable)
self.assertIsInstance(mb_confs.data, type(xs[0]))
self.assertEqual(mb_confs.shape,
(self.batchsize, self.n_bbox, self.n_class))
self.assertEqual(mb_confs.dtype, xs[0].dtype)
def test_forward_cpu(self):
self._check_forward(self.xs)
@attr.gpu
def test_forward_gpu(self):
self.link.to_gpu()
self._check_forward(list(map(chainer.cuda.to_gpu, self.xs)))
def __init__(self, n_fg_class=None,
pretrained_extractor='auto',
pretrained_model=None):
super(SSD224, self).__init__(
extractor=ResNet101Extractor(pretrained_extractor),
multibox=Multibox(
n_class=n_fg_class + 1,
aspect_ratios=((2, 3), (2, 3), (2, 3), (2, 3))),
steps=(4, 8, 16, 32),
sizes=(15, 30, 60, 120, 244),
mean=_imagenet_mean)
if pretrained_model:
_load_npz(pretrained_model, self)
def __init__(self, n_fg_class=None, pretrained_model=None):
param, path = utils.prepare_pretrained_model(
{'n_fg_class': n_fg_class}, pretrained_model, self._models)
super(M2Det300, self).__init__(
extractor=VGG16MLFPN300(),
multibox=Multibox(
n_class=param['n_fg_class'] + 1,
aspect_ratios=((2,), (2, 3), (2, 3), (2, 3), (2,), (2,))),
steps=(8, 16, 32, 64, 100, 300),
sizes=(30, 60, 111, 162, 213, 264, 315),
mean=_imagenet_mean)
if path:
chainer.serializers.load_npz(path, self, strict=False)
def __init__(self, n_fg_class=None, pretrained_model=None):
param, path = utils.prepare_pretrained_model(
{'n_fg_class': n_fg_class}, pretrained_model, self._models)
super(M2Det512, self).__init__(
extractor=VGG16MLFPN512(),
multibox=Multibox(
n_class=param['n_fg_class'] + 1,
aspect_ratios=((2,), (2, 3), (2, 3), (2, 3), (2,), (2,))),
steps=(8, 16, 32, 64, 128, 256),
sizes=(30.72, 76.8, 168.96, 261.12, 353.28, 445.44, 537.6),
mean=_imagenet_mean)
if path:
chainer.serializers.load_npz(path, self, strict=False)
def __init__(self, n_fg_class=None, pretrained_model=None):
param, path = utils.prepare_pretrained_model(
{'n_fg_class': n_fg_class}, pretrained_model, self._models)
super(M2Det320, self).__init__(
extractor=VGG16MLFPN320(),
multibox=Multibox(
n_class=param['n_fg_class'] + 1,
aspect_ratios=((2,), (2, 3), (2, 3), (2, 3), (2,), (2,))),
steps=(8, 16, 32, 64, 107, 320),
sizes=(25.6, 48, 105.6, 163.2, 220.8, 278.4, 336),
mean=_imagenet_mean)
if path:
chainer.serializers.load_npz(path, self, strict=False)
def __init__(self, n_fg_class=None, pretrained_model=None):
param, path = utils.prepare_pretrained_model(
{'n_fg_class': n_fg_class}, pretrained_model, self._models)
super(SSD512, self).__init__(
extractor=VGG16Extractor512(),
multibox=Multibox(n_class=param['n_fg_class'] + 1,
aspect_ratios=((2, ), (2, 3), (2, 3), (2, 3),
(2, 3), (2, ), (2, ))),
steps=(8, 16, 32, 64, 128, 256, 512),
sizes=(35.84, 76.8, 153.6, 230.4, 307.2, 384.0, 460.8, 537.6),
mean=_imagenet_mean)
if path:
chainer.serializers.load_npz(path, self, strict=False)
def __init__(self, n_fg_class=None, pretrained_model=None):
n_fg_class, path = _check_pretrained_model(
n_fg_class, pretrained_model, self._models)
super(SSD300, self).__init__(
extractor=VGG16Extractor300(),
multibox=Multibox(
n_class=n_fg_class + 1,
aspect_ratios=((2,), (2, 3), (2, 3), (2, 3), (2,), (2,))),
steps=[s / 300 for s in (8, 16, 32, 64, 100, 300)],
sizes=[s / 300 for s in (30, 60, 111, 162, 213, 264, 315)],
mean=_imagenet_mean)
if path:
chainer.serializers.load_npz(path, self)
def __init__(self, n_fg_class=None, pretrained_model=None):
n_fg_class, path = _check_pretrained_model(n_fg_class,
pretrained_model,
self._models)
super(SSD512, self).__init__(
extractor=VGG16Extractor512(),
multibox=Multibox(n_class=n_fg_class + 1,
aspect_ratios=((2, ), (2, 3), (2, 3), (2, 3),
(2, 3), (2, ), (2, ))),
steps=(8, 16, 32, 64, 128, 256, 512),
sizes=(35.84, 76.8, 153.6, 230.4, 307.2, 384.0, 460.8, 537.6),
mean=_imagenet_mean)
if path:
_load_npz(path, self)
def __init__(self,
n_fg_class=None,
pretrained_model=None,
init_scale=0.0001):
# うまくいったらpretrained model配布しますね
# n_fg_class, path = _check_pretrained_model(
# n_fg_class, pretrained_model, self._models)
print('initializing with scale={}'.format(init_scale))
initializer = initializers.Normal(init_scale)
super(FPNSSD, self).__init__(
extractor=FeaturePyramidNetwork(initialW=initializer),
multibox=Multibox(n_class=n_fg_class + 1,
aspect_ratios=((2, ), (2, 3), (2, 3), (2, 3),
(2, ), (2, )),
initialW=initializer),
steps=(8, 16, 32, 64, 100, 300),
sizes=(30, 60, 111, 162, 213, 264, 315),
mean=_imagenet_mean)
