Spaces:
Running
on
Zero
Running
on
Zero
File size: 10,943 Bytes
4893ce0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 |
from functools import partial
import torch
import torch.nn as nn
from einops import rearrange
import spconv.pytorch as spconv
from timm.models.layers import trunc_normal_
from ..builder import MODELS
from ..utils import offset2batch
from torch_geometric.nn.pool import voxel_grid
from torch_geometric.utils import scatter
class BasicBlock(nn.Module):
def __init__(
self,
in_channels,
embed_channels,
norm_fn=None,
indice_key=None,
depth=4,
groups=None,
grid_size=None,
bias=False,
):
super().__init__()
assert embed_channels % groups == 0
self.groups = groups
self.embed_channels = embed_channels
self.proj = nn.ModuleList()
self.grid_size = grid_size
self.weight = nn.ModuleList()
self.l_w = nn.ModuleList()
self.proj.append(
nn.Sequential(
nn.Linear(embed_channels, embed_channels, bias=False),
norm_fn(embed_channels),
nn.ReLU(),
)
)
for _ in range(depth - 1):
self.proj.append(
nn.Sequential(
nn.Linear(embed_channels, embed_channels, bias=False),
norm_fn(embed_channels),
nn.ReLU(),
)
)
self.l_w.append(
nn.Sequential(
nn.Linear(embed_channels, embed_channels, bias=False),
norm_fn(embed_channels),
nn.ReLU(),
)
)
self.weight.append(nn.Linear(embed_channels, embed_channels, bias=False))
self.adaptive = nn.Linear(embed_channels, depth - 1, bias=False)
self.fuse = nn.Sequential(
nn.Linear(embed_channels * 2, embed_channels, bias=False),
norm_fn(embed_channels),
nn.ReLU(),
)
self.voxel_block = spconv.SparseSequential(
spconv.SubMConv3d(
embed_channels,
embed_channels,
kernel_size=3,
stride=1,
padding=1,
indice_key=indice_key,
bias=bias,
),
norm_fn(embed_channels),
nn.ReLU(),
spconv.SubMConv3d(
embed_channels,
embed_channels,
kernel_size=3,
stride=1,
padding=1,
indice_key=indice_key,
bias=bias,
),
norm_fn(embed_channels),
)
self.act = nn.ReLU()
def forward(self, x, clusters):
feat = x.features
feats = []
for i, cluster in enumerate(clusters):
pw = self.l_w[i](feat)
pw = pw - scatter(pw, cluster, reduce="mean")[cluster]
pw = self.weight[i](pw)
pw = torch.exp(pw - pw.max())
pw = pw / (scatter(pw, cluster, reduce="sum", dim=0)[cluster] + 1e-6)
pfeat = self.proj[i](feat) * pw
pfeat = scatter(pfeat, cluster, reduce="sum")[cluster]
feats.append(pfeat)
adp = self.adaptive(feat)
adp = torch.softmax(adp, dim=1)
feats = torch.stack(feats, dim=1)
feats = torch.einsum("l n, l n c -> l c", adp, feats)
feat = self.proj[-1](feat)
feat = torch.cat([feat, feats], dim=1)
feat = self.fuse(feat) + x.features
res = feat
x = x.replace_feature(feat)
x = self.voxel_block(x)
x = x.replace_feature(self.act(x.features + res))
return x
class DonwBlock(nn.Module):
def __init__(
self,
in_channels,
embed_channels,
depth,
sp_indice_key,
point_grid_size,
num_ref=16,
groups=None,
norm_fn=None,
sub_indice_key=None,
):
super().__init__()
self.num_ref = num_ref
self.depth = depth
self.point_grid_size = point_grid_size
self.down = spconv.SparseSequential(
spconv.SparseConv3d(
in_channels,
embed_channels,
kernel_size=2,
stride=2,
indice_key=sp_indice_key,
bias=False,
),
norm_fn(embed_channels),
nn.ReLU(),
)
self.blocks = nn.ModuleList()
for _ in range(depth):
self.blocks.append(
BasicBlock(
in_channels=embed_channels,
embed_channels=embed_channels,
depth=len(point_grid_size) + 1,
groups=groups,
grid_size=point_grid_size,
norm_fn=norm_fn,
indice_key=sub_indice_key,
)
)
def forward(self, x):
x = self.down(x)
coord = x.indices[:, 1:].float()
batch = x.indices[:, 0]
clusters = []
for grid_size in self.point_grid_size:
cluster = voxel_grid(pos=coord, size=grid_size, batch=batch)
_, cluster = torch.unique(cluster, return_inverse=True)
clusters.append(cluster)
for block in self.blocks:
x = block(x, clusters)
return x
class UpBlock(nn.Module):
def __init__(
self,
in_channels,
skip_channels,
embed_channels,
depth,
sp_indice_key,
norm_fn=None,
down_ratio=2,
sub_indice_key=None,
):
super().__init__()
assert depth > 0
self.up = spconv.SparseSequential(
spconv.SparseInverseConv3d(
in_channels,
embed_channels,
kernel_size=down_ratio,
indice_key=sp_indice_key,
bias=False,
),
norm_fn(embed_channels),
nn.ReLU(),
)
self.blocks = nn.ModuleList()
self.fuse = nn.Sequential(
nn.Linear(skip_channels + embed_channels, embed_channels),
norm_fn(embed_channels),
nn.ReLU(),
nn.Linear(embed_channels, embed_channels),
norm_fn(embed_channels),
nn.ReLU(),
)
def forward(self, x, skip_x):
x = self.up(x)
x = x.replace_feature(
self.fuse(torch.cat([x.features, skip_x.features], dim=1)) + x.features
)
return x
@MODELS.register_module()
class OACNNs(nn.Module):
def __init__(
self,
in_channels,
num_classes,
embed_channels=64,
enc_num_ref=[16, 16, 16, 16],
enc_channels=[64, 64, 128, 256],
groups=[2, 4, 8, 16],
enc_depth=[2, 3, 6, 4],
down_ratio=[2, 2, 2, 2],
dec_channels=[96, 96, 128, 256],
point_grid_size=[[16, 32, 64], [8, 16, 24], [4, 8, 12], [2, 4, 6]],
dec_depth=[2, 2, 2, 2],
):
super().__init__()
self.in_channels = in_channels
self.num_classes = num_classes
self.num_stages = len(enc_channels)
self.embed_channels = embed_channels
norm_fn = partial(nn.BatchNorm1d, eps=1e-3, momentum=0.01)
self.stem = spconv.SparseSequential(
spconv.SubMConv3d(
in_channels,
embed_channels,
kernel_size=3,
padding=1,
indice_key="stem",
bias=False,
),
norm_fn(embed_channels),
nn.ReLU(),
spconv.SubMConv3d(
embed_channels,
embed_channels,
kernel_size=3,
padding=1,
indice_key="stem",
bias=False,
),
norm_fn(embed_channels),
nn.ReLU(),
spconv.SubMConv3d(
embed_channels,
embed_channels,
kernel_size=3,
padding=1,
indice_key="stem",
bias=False,
),
norm_fn(embed_channels),
nn.ReLU(),
)
self.enc = nn.ModuleList()
self.dec = nn.ModuleList()
for i in range(self.num_stages):
self.enc.append(
DonwBlock(
in_channels=embed_channels if i == 0 else enc_channels[i - 1],
embed_channels=enc_channels[i],
depth=enc_depth[i],
norm_fn=norm_fn,
groups=groups[i],
point_grid_size=point_grid_size[i],
num_ref=enc_num_ref[i],
sp_indice_key=f"spconv{i}",
sub_indice_key=f"subm{i + 1}",
)
)
self.dec.append(
UpBlock(
in_channels=(
enc_channels[-1]
if i == self.num_stages - 1
else dec_channels[i + 1]
),
skip_channels=embed_channels if i == 0 else enc_channels[i - 1],
embed_channels=dec_channels[i],
depth=dec_depth[i],
norm_fn=norm_fn,
sp_indice_key=f"spconv{i}",
sub_indice_key=f"subm{i}",
)
)
self.final = spconv.SubMConv3d(dec_channels[0], num_classes, kernel_size=1)
self.apply(self._init_weights)
def forward(self, input_dict):
discrete_coord = input_dict["grid_coord"]
feat = input_dict["feat"]
offset = input_dict["offset"]
batch = offset2batch(offset)
x = spconv.SparseConvTensor(
features=feat,
indices=torch.cat([batch.unsqueeze(-1), discrete_coord], dim=1)
.int()
.contiguous(),
spatial_shape=torch.add(
torch.max(discrete_coord, dim=0).values, 1
).tolist(),
batch_size=batch[-1].tolist() + 1,
)
x = self.stem(x)
skips = [x]
for i in range(self.num_stages):
x = self.enc[i](x)
skips.append(x)
x = skips.pop(-1)
for i in reversed(range(self.num_stages)):
skip = skips.pop(-1)
x = self.dec[i](x, skip)
x = self.final(x)
return x.features
@staticmethod
def _init_weights(m):
if isinstance(m, nn.Linear):
trunc_normal_(m.weight, std=0.02)
if m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, spconv.SubMConv3d):
trunc_normal_(m.weight, std=0.02)
if m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm1d):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
|