def test_chamfer_is_better_after_fit(self):
d = load_test_files()
dist_orig = metrics.chamfer_distance(d[0], d[1], scale=0.1)
vg_fit = fit.icp(d[1], d[0], scale=0.1)
dist_fit = metrics.chamfer_distance(d[0], vg_fit, scale=0.1)
self.assertLess(dist_fit, dist_orig,
"Distance was not lower after fitting with icp")
def test_chamfer_distance_voxelgrid_on_simple(self):
vg1 = tf.cast([[[0, 0, 0], [0, 1, 0], [0, 1, 0]],
[[0, 0, 0], [0, 0, 0], [0, 0, 0]],
[[0, 0, 0], [0, 0, 0], [0, 0, 0]]], tf.float32)
vg2 = tf.cast([[[0, 0, 0], [0, 1, 0], [0, 0, 0]],
[[0, 0, 0], [0, 1, 0], [0, 0, 0]],
[[0, 0, 0], [0, 0, 0], [0, 0, 0]]], tf.float32)
d = (0.0 + 1)/2 + (0.0 + 1)/2
self.assertEqual(d, metrics.chamfer_distance(vg1, vg2, scale=1.0))
self.assertEqual(d/4, metrics.chamfer_distance(vg1, vg2, scale=1.0/4))
def run_inference(elem):
if ARGS.enforce_contact:
return wip_enforce_contact(elem)
if ARGS.publish_closest_train:
# Computes and publishes the closest element in the training set to the test shape
train_in_correct_augmentation = train_records.filter(lambda x: x['augmentation'] == elem['augmentation'][0])
train_in_correct_augmentation = shape_completion_training.utils.old_dataset_tools.load_voxelgrids(train_in_correct_augmentation)
min_cd = np.inf
closest_train = None
for train_elem in train_in_correct_augmentation:
VG_PUB.publish("plausible", train_elem['gt_occ'])
cd = chamfer_distance(elem['gt_occ'], train_elem['gt_occ'],
scale=0.01, downsample=4)
if cd < min_cd:
min_cd = cd
closest_train = train_elem['gt_occ']
VG_PUB.publish("plausible", closest_train)
# raw_input("Ready to display best?")
inference = model_runner.model(elem)
VG_PUB.publish_inference(inference)
return inference
def test_chamfer_distance_voxelgrid_on_real(self):
d = load_test_files()
self.assertEqual(0.0, metrics.chamfer_distance(d[0], d[0], scale=0.1))
self.assertGreater(metrics.chamfer_distance(d[0], d[1], scale=0.1),
0.0)
def m(vg1, vg2):
vg_fit = fit.icp(vg2, vg1, scale=0.1, max_iter=10, downsample=2)
return metrics.chamfer_distance(vg1,
vg_fit,
scale=0.1,
downsample=2)
def vg_chamfer_distance(vg1, vg2, scale):
return chamfer_distance(vg1, vg2, scale, downsample=3)