def forward(self, data, octree):
assert data.size(1) == self.channel_in
conv = octree_conv(
data, self.weights, octree, self.depth, self.channel_out,
self.kernel_size, self.stride, self.nempty)
if self.stride == 2 and not self.nempty:
conv = ocnn.octree_pad(conv, octree, self.depth-1)
return conv
def forward(self, data, octree):
depth = self.depth
col = ocnn.octree2col(data, octree, depth, self.kernel_size, self.stride, False)
col = col.view([self.cdim, -1])
conv = torch.mm(self.weights, col)
conv = torch.unsqueeze(torch.unsqueeze(conv, 0), -1) # [C,H] -> [1,C,H,1]
if self.stride == 2:
conv = ocnn.octree_pad(conv, octree, depth-1)
return conv
def test_octree2colP(self):
depth = 4
channel = 5
stride = [1, 2]
kernel_size = [[3, 3, 3], [2, 2, 2], [3, 1, 1], [3, 3, 1], [1, 1, 1]]
samples = ocnn.octree_samples(
['octree_1', 'octree_2', 'octree_2', 'octree_1'])
octree = ocnn.octree_batch(samples).cuda()
node_num = ocnn.octree_property(octree, 'node_num', depth)
data_in = torch.rand(1, channel, node_num.item(), 1).cuda()
data_in = ocnn.octree_depad(data_in, octree, depth)
data_in1 = data_in.clone().requires_grad_()
data1 = ocnn.octree_pad(data_in1, octree, depth, 0)
data_in2 = data_in.clone().requires_grad_()
# octree2colP = octree2col + depad
for i in range(len(stride)):
for j in range(len(kernel_size)):
out1 = ocnn.octree2col(data1, octree, depth, kernel_size[j],
stride[i], False)
if stride[i] == 1:
ks, height = out1.size(1), out1.size(2)
out1 = out1.view(1, -1, height, 1)
out1 = ocnn.octree_depad(out1, octree, depth)
out1 = out1.view(channel, ks, -1)
out2 = ocnn.octree2col(data_in2, octree, depth, kernel_size[j],
stride[i], True)
pesudo_grad = torch.rand(out1.shape,
dtype=out1.dtype,
device=out1.device)
out1.backward(pesudo_grad, retain_graph=True)
out2.backward(pesudo_grad, retain_graph=True)
# check
self.assertTrue(
np.array_equal(out1.detach().cpu().numpy(),
out2.detach().cpu().numpy()))
self.assertTrue(
np.allclose(data_in1.grad.cpu().numpy(),
data_in2.grad.cpu().numpy()))
def forward(self, data, octree): channel = data.shape[1] data = data.view(1, channel, -1, 8) pool = data.mean(dim=3, keepdims=True) output = ocnn.octree_pad(pool, octree, self.depth-1) # !!! depth-1 return output
def forward(self, data, octree): pool, mask = octree_max_pool(data, octree, self.depth) output = ocnn.octree_pad(pool, octree, self.depth-1) # !!! depth-1 return output if not self.return_indices else (output, mask)
def forward(self, data_in, octree):
conv = octree_conv(data_in, self.weights, octree, self.depth,
self.channel_out, self.kernel_size, self.stride)
if self.stride == 2:
conv = ocnn.octree_pad(conv, octree, self.depth - 1)
return conv