def forward(self, data):
if self._backbone_name == "kpconv": # KPConv needs added ones for its x features
data = AddOnes()(data)
data.x = data.ones
data_out = self._model["backbone"](data)
return F.log_softmax(self._model["classifier"](data_out.x).squeeze(),
dim=-1)
def set_input(self, data, device):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input: a dictionary that contains the data itself and its metadata information.
"""
self.semantic_labels = data.y.flatten().to(self.device)
if self.is_dense_format:
self.input = data
self._n_batches = data.pos.shape[0]
else:
if self._kpconv_backbone:
data = AddOnes()(data)
if data.x is not None:
data.x = torch.cat([data.x, data.ones.float()], dim=-1)
else:
data.x = data.ones.float()
self.input = data
self._n_batches = torch.max(data.batch) + 1
def _set_input(self, data):
"""Unpack input data from the dataloader and perform necessary pre-processing steps.
Parameters:
input: a dictionary that contains the data itself and its metadata information.
"""
# Forward through backbone model
if self.is_dense_format:
self.input = data
self._num_batches = data.pos.shape[0]
else:
if self._kpconv_backbone:
data = AddOnes()(data)
if data.x is not None:
data.x = torch.cat([data.x, data.ones.float()], dim=-1)
else:
data.x = data.ones.float()
self.input = Data(pos=data.pos, x=data.x,
batch=data.batch).to(self.device)
self._num_batches = len(data.id_scan)
-6.481465826011e-03, 8.051860151134e-03, -9.999466081774e-01,
-7.337429464231e-02
],
[
9.999773098287e-01, -1.805528627661e-03, -6.496203536139e-03,
-3.339968064433e-01
],
])
pcd_s = np.fromfile(path_s, dtype=np.float32).reshape(-1, 4)[:, :3].dot(
R_calib[:3, :3].T)
pcd_t = np.fromfile(path_t, dtype=np.float32).reshape(-1, 4)[:, :3].dot(
R_calib[:3, :3].T)
transform = Compose([
GridSampling3D(mode="last", size=0.3, quantize_coords=True),
AddOnes(),
AddFeatByKey(add_to_x=True, feat_name="ones"),
])
data_s = transform(
Batch(pos=torch.from_numpy(pcd_s).float(),
batch=torch.zeros(pcd_s.shape[0]).long()))
data_t = transform(
Batch(pos=torch.from_numpy(pcd_t).float(),
batch=torch.zeros(pcd_t.shape[0]).long()))
model = PretainedRegistry.from_pretrained(
"minkowski-registration-kitti").cuda()
with torch.no_grad():
model.set_input(data_s, "cuda")