def __init__(self, n_fg_class):
extractor = DummyExtractor()
super(DummyFasterRCNN, self).__init__(
extractor=extractor,
rpn=RPN(extractor.scales),
head=Head(n_fg_class + 1, extractor.scales),
)
def _check_rpn_loss(self, xp):
locs = [
chainer.Variable(_random_array(xp, (2, 32 * 32 * 3, 4))),
chainer.Variable(_random_array(xp, (2, 16 * 16 * 3, 4))),
chainer.Variable(_random_array(xp, (2, 8 * 8 * 3, 4))),
]
confs = [
chainer.Variable(_random_array(xp, (2, 32 * 32 * 3))),
chainer.Variable(_random_array(xp, (2, 16 * 16 * 3))),
chainer.Variable(_random_array(xp, (2, 8 * 8 * 3))),
]
anchors = RPN(scales=(1 / 2, 1 / 4, 1 / 8)) \
.anchors(((32, 32), (16, 16), (8, 8)))
bboxes = [
xp.array(((2, 4, 6, 7), (1, 12, 3, 30)), dtype=np.float32),
xp.array(((10, 2, 12, 12), ), dtype=np.float32),
]
loc_loss, conf_loss = rpn_loss(locs, confs, anchors,
((480, 640), (320, 320)), bboxes)
self.assertIsInstance(loc_loss, chainer.Variable)
self.assertIsInstance(loc_loss.array, xp.ndarray)
self.assertEqual(loc_loss.shape, ())
self.assertIsInstance(conf_loss, chainer.Variable)
self.assertIsInstance(conf_loss.array, xp.ndarray)
self.assertEqual(conf_loss.shape, ())
def __init__(self, n_fg_class, return_values, min_size, max_size):
extractor = DummyExtractor()
super(DummyFasterRCNN, self).__init__(
extractor=extractor,
rpn=RPN(extractor.scales),
bbox_head=BboxHead(n_fg_class + 1, extractor.scales),
mask_head=MaskHead(n_fg_class + 1, extractor.scales),
return_values=return_values,
min_size=min_size, max_size=max_size,
)
def setUp(self):
self.link = RPN(scales=(1 / 2, 1 / 4, 1 / 8))
class TestRPN(unittest.TestCase):
def setUp(self):
self.link = RPN(scales=(1 / 2, 1 / 4, 1 / 8))
def _check_call(self):
hs = [
chainer.Variable(_random_array(self.link.xp, (2, 64, 32, 32))),
chainer.Variable(_random_array(self.link.xp, (2, 64, 16, 16))),
chainer.Variable(_random_array(self.link.xp, (2, 64, 8, 8))),
]
locs, confs = self.link(hs)
self.assertEqual(len(locs), 3)
self.assertEqual(len(confs), 3)
for l in range(3):
self.assertIsInstance(locs[l], chainer.Variable)
self.assertIsInstance(locs[l].array, self.link.xp.ndarray)
self.assertEqual(locs[l].shape, (2, (32 * 32 >> 2 * l) * 3, 4))
self.assertIsInstance(confs[l], chainer.Variable)
self.assertIsInstance(confs[l].array, self.link.xp.ndarray)
self.assertEqual(confs[l].shape, (2, (32 * 32 >> 2 * l) * 3))
def test_call_cpu(self):
self._check_call()
@attr.gpu
def test_call_gpu(self):
self.link.to_gpu()
self._check_call()
def _check_anchors(self):
anchors = self.link.anchors(((32, 32), (16, 16), (8, 8)))
self.assertEqual(len(anchors), 3)
for l in range(3):
self.assertIsInstance(anchors[l], self.link.xp.ndarray)
self.assertEqual(anchors[l].shape, ((32 * 32 >> 2 * l) * 3, 4))
def test_anchors_cpu(self):
self._check_anchors()
@attr.gpu
def test_anchors_gpu(self):
self.link.to_gpu()
self._check_anchors()
def _check_decode(self):
locs = [
chainer.Variable(_random_array(self.link.xp, (2, 32 * 32 * 3, 4))),
chainer.Variable(_random_array(self.link.xp, (2, 16 * 16 * 3, 4))),
chainer.Variable(_random_array(self.link.xp, (2, 8 * 8 * 3, 4))),
]
confs = [
chainer.Variable(_random_array(self.link.xp, (2, 32 * 32 * 3))),
chainer.Variable(_random_array(self.link.xp, (2, 16 * 16 * 3))),
chainer.Variable(_random_array(self.link.xp, (2, 8 * 8 * 3))),
]
anchors = self.link.anchors(((32, 32), (16, 16), (8, 8)))
rois, roi_indices = self.link.decode(locs, confs, anchors,
(2, 3, 64, 64))
self.assertIsInstance(rois, self.link.xp.ndarray)
self.assertIsInstance(roi_indices, self.link.xp.ndarray)
self.assertEqual(rois.shape[0], roi_indices.shape[0])
self.assertEqual(rois.shape[1:], (4, ))
self.assertEqual(roi_indices.shape[1:], ())
def test_decode_cpu(self):
self._check_decode()
@attr.gpu
def test_decode_gpu(self):
self.link.to_gpu()
self._check_decode()