resnet.py 17.9 KB
Newer Older
Q
qingqing01 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. 
#   
# Licensed under the Apache License, Version 2.0 (the "License");   
# you may not use this file except in compliance with the License.  
# You may obtain a copy of the License at   
#   
#     http://www.apache.org/licenses/LICENSE-2.0    
#   
# Unless required by applicable law or agreed to in writing, software   
# distributed under the License is distributed on an "AS IS" BASIS, 
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.  
# See the License for the specific language governing permissions and   
# limitations under the License.

15
import math
16 17
from numbers import Integral

Q
qingqing01 已提交
18 19 20 21 22
import paddle
import paddle.nn as nn
import paddle.nn.functional as F
from ppdet.core.workspace import register, serializable
from paddle.regularizer import L2Decay
F
Feng Ni 已提交
23
from ppdet.modeling.layers import DeformableConvV2
24 25
from .name_adapter import NameAdapter
from ..shape_spec import ShapeSpec
Q
qingqing01 已提交
26

27 28
__all__ = ['ResNet', 'Res5Head']

29 30 31 32 33 34 35 36
ResNet_cfg = {
    18: [2, 2, 2, 2],
    34: [3, 4, 6, 3],
    50: [3, 4, 6, 3],
    101: [3, 4, 23, 3],
    152: [3, 8, 36, 3],
}

Q
qingqing01 已提交
37 38 39 40 41 42 43 44

class ConvNormLayer(nn.Layer):
    def __init__(self,
                 ch_in,
                 ch_out,
                 filter_size,
                 stride,
                 name_adapter,
45
                 groups=1,
Q
qingqing01 已提交
46 47 48 49 50
                 act=None,
                 norm_type='bn',
                 norm_decay=0.,
                 freeze_norm=True,
                 lr=1.0,
F
Feng Ni 已提交
51
                 dcn_v2=False,
Q
qingqing01 已提交
52 53 54 55 56 57
                 name=None):
        super(ConvNormLayer, self).__init__()
        assert norm_type in ['bn', 'sync_bn']
        self.norm_type = norm_type
        self.act = act

F
Feng Ni 已提交
58
        if not dcn_v2:
59
            self.conv = nn.Conv2D(
F
Feng Ni 已提交
60 61 62 63 64
                in_channels=ch_in,
                out_channels=ch_out,
                kernel_size=filter_size,
                stride=stride,
                padding=(filter_size - 1) // 2,
65
                groups=groups,
66
                weight_attr=paddle.ParamAttr(
F
Feng Ni 已提交
67 68 69 70 71 72 73 74 75
                    learning_rate=lr, name=name + "_weights"),
                bias_attr=False)
        else:
            self.conv = DeformableConvV2(
                in_channels=ch_in,
                out_channels=ch_out,
                kernel_size=filter_size,
                stride=stride,
                padding=(filter_size - 1) // 2,
76
                groups=groups,
77 78
                weight_attr=paddle.ParamAttr(
                    learning_rate=lr, name=name + '_weights'),
F
Feng Ni 已提交
79 80
                bias_attr=False,
                name=name)
Q
qingqing01 已提交
81 82 83

        bn_name = name_adapter.fix_conv_norm_name(name)
        norm_lr = 0. if freeze_norm else lr
84
        param_attr = paddle.ParamAttr(
Q
qingqing01 已提交
85 86 87 88
            learning_rate=norm_lr,
            regularizer=L2Decay(norm_decay),
            name=bn_name + "_scale",
            trainable=False if freeze_norm else True)
89
        bias_attr = paddle.ParamAttr(
Q
qingqing01 已提交
90 91 92 93 94 95
            learning_rate=norm_lr,
            regularizer=L2Decay(norm_decay),
            name=bn_name + "_offset",
            trainable=False if freeze_norm else True)

        global_stats = True if freeze_norm else False
96 97 98 99 100 101 102 103 104 105 106 107
        if norm_type == 'sync_bn':
            self.norm = nn.SyncBatchNorm(
                ch_out, weight_attr=param_attr, bias_attr=bias_attr)
        else:
            self.norm = nn.BatchNorm(
                ch_out,
                act=None,
                param_attr=param_attr,
                bias_attr=bias_attr,
                use_global_stats=global_stats,
                moving_mean_name=bn_name + '_mean',
                moving_variance_name=bn_name + '_variance')
Q
qingqing01 已提交
108 109 110 111 112 113 114 115
        norm_params = self.norm.parameters()

        if freeze_norm:
            for param in norm_params:
                param.stop_gradient = True

    def forward(self, inputs):
        out = self.conv(inputs)
116
        if self.norm_type in ['bn', 'sync_bn']:
Q
qingqing01 已提交
117
            out = self.norm(out)
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
        if self.act:
            out = getattr(F, self.act)(out)
        return out


class BasicBlock(nn.Layer):
    def __init__(self,
                 ch_in,
                 ch_out,
                 stride,
                 shortcut,
                 name_adapter,
                 name,
                 variant='b',
                 lr=1.0,
                 norm_type='bn',
                 norm_decay=0.,
                 freeze_norm=True,
                 dcn_v2=False):
        super(BasicBlock, self).__init__()
        assert dcn_v2 is False, "Not implemented yet."
        conv_name1, conv_name2, shortcut_name = name_adapter.fix_basicblock_name(
            name)

        self.shortcut = shortcut
        if not shortcut:
            if variant == 'd' and stride == 2:
                self.short = nn.Sequential()
                self.short.add_sublayer(
                    'pool',
                    nn.AvgPool2D(
                        kernel_size=2, stride=2, padding=0, ceil_mode=True))
                self.short.add_sublayer(
                    'conv',
                    ConvNormLayer(
                        ch_in=ch_in,
                        ch_out=ch_out,
                        filter_size=1,
                        stride=1,
                        name_adapter=name_adapter,
                        norm_type=norm_type,
                        norm_decay=norm_decay,
                        freeze_norm=freeze_norm,
                        lr=lr,
                        name=shortcut_name))
            else:
                self.short = ConvNormLayer(
                    ch_in=ch_in,
                    ch_out=ch_out,
                    filter_size=1,
                    stride=stride,
                    name_adapter=name_adapter,
                    norm_type=norm_type,
                    norm_decay=norm_decay,
                    freeze_norm=freeze_norm,
                    lr=lr,
                    name=shortcut_name)

        self.branch2a = ConvNormLayer(
            ch_in=ch_in,
            ch_out=ch_out,
            filter_size=3,
            stride=stride,
            name_adapter=name_adapter,
            act='relu',
            norm_type=norm_type,
            norm_decay=norm_decay,
            freeze_norm=freeze_norm,
            lr=lr,
            name=conv_name1)

        self.branch2b = ConvNormLayer(
            ch_in=ch_out,
            ch_out=ch_out,
            filter_size=3,
            stride=1,
            name_adapter=name_adapter,
            act=None,
            norm_type=norm_type,
            norm_decay=norm_decay,
            freeze_norm=freeze_norm,
            lr=lr,
            name=conv_name2)

    def forward(self, inputs):
        out = self.branch2a(inputs)
        out = self.branch2b(out)

        if self.shortcut:
            short = inputs
        else:
            short = self.short(inputs)

        out = paddle.add(x=out, y=short)
        out = F.relu(out)

Q
qingqing01 已提交
214 215 216 217 218 219 220 221 222 223 224 225
        return out


class BottleNeck(nn.Layer):
    def __init__(self,
                 ch_in,
                 ch_out,
                 stride,
                 shortcut,
                 name_adapter,
                 name,
                 variant='b',
226 227 228
                 groups=1,
                 base_width=4,
                 base_channels=64,
Q
qingqing01 已提交
229 230 231
                 lr=1.0,
                 norm_type='bn',
                 norm_decay=0.,
F
Feng Ni 已提交
232 233
                 freeze_norm=True,
                 dcn_v2=False):
Q
qingqing01 已提交
234 235 236 237 238 239
        super(BottleNeck, self).__init__()
        if variant == 'a':
            stride1, stride2 = stride, 1
        else:
            stride1, stride2 = 1, stride

240 241 242 243 244 245 246 247
        # ResNeXt
        if groups == 1:
            width = ch_out
        else:
            width = int(
                math.floor(ch_out * (base_width * 1.0 / base_channels)) *
                groups)

Q
qingqing01 已提交
248 249 250 251 252
        conv_name1, conv_name2, conv_name3, \
            shortcut_name = name_adapter.fix_bottleneck_name(name)

        self.shortcut = shortcut
        if not shortcut:
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
            if variant == 'd' and stride == 2:
                self.short = nn.Sequential()
                self.short.add_sublayer(
                    'pool',
                    nn.AvgPool2D(
                        kernel_size=2, stride=2, padding=0, ceil_mode=True))
                self.short.add_sublayer(
                    'conv',
                    ConvNormLayer(
                        ch_in=ch_in,
                        ch_out=ch_out * 4,
                        filter_size=1,
                        stride=1,
                        name_adapter=name_adapter,
                        norm_type=norm_type,
                        norm_decay=norm_decay,
                        freeze_norm=freeze_norm,
                        lr=lr,
                        name=shortcut_name))
            else:
                self.short = ConvNormLayer(
                    ch_in=ch_in,
                    ch_out=ch_out * 4,
                    filter_size=1,
                    stride=stride,
                    name_adapter=name_adapter,
                    norm_type=norm_type,
                    norm_decay=norm_decay,
                    freeze_norm=freeze_norm,
                    lr=lr,
                    name=shortcut_name)
Q
qingqing01 已提交
284 285 286

        self.branch2a = ConvNormLayer(
            ch_in=ch_in,
287
            ch_out=width,
Q
qingqing01 已提交
288 289 290
            filter_size=1,
            stride=stride1,
            name_adapter=name_adapter,
291
            groups=1,
Q
qingqing01 已提交
292 293 294 295 296 297 298 299
            act='relu',
            norm_type=norm_type,
            norm_decay=norm_decay,
            freeze_norm=freeze_norm,
            lr=lr,
            name=conv_name1)

        self.branch2b = ConvNormLayer(
300 301
            ch_in=width,
            ch_out=width,
Q
qingqing01 已提交
302 303 304
            filter_size=3,
            stride=stride2,
            name_adapter=name_adapter,
305
            groups=groups,
Q
qingqing01 已提交
306 307 308 309 310
            act='relu',
            norm_type=norm_type,
            norm_decay=norm_decay,
            freeze_norm=freeze_norm,
            lr=lr,
F
Feng Ni 已提交
311
            dcn_v2=dcn_v2,
Q
qingqing01 已提交
312 313 314
            name=conv_name2)

        self.branch2c = ConvNormLayer(
315
            ch_in=width,
Q
qingqing01 已提交
316 317 318 319
            ch_out=ch_out * 4,
            filter_size=1,
            stride=1,
            name_adapter=name_adapter,
320
            groups=1,
Q
qingqing01 已提交
321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344
            norm_type=norm_type,
            norm_decay=norm_decay,
            freeze_norm=freeze_norm,
            lr=lr,
            name=conv_name3)

    def forward(self, inputs):

        out = self.branch2a(inputs)
        out = self.branch2b(out)
        out = self.branch2c(out)
        if self.shortcut:
            short = inputs
        else:
            short = self.short(inputs)

        out = paddle.add(x=out, y=short)
        out = F.relu(out)

        return out


class Blocks(nn.Layer):
    def __init__(self,
345
                 depth,
Q
qingqing01 已提交
346 347 348 349 350
                 ch_in,
                 ch_out,
                 count,
                 name_adapter,
                 stage_num,
351 352 353 354
                 variant='b',
                 groups=1,
                 base_width=-1,
                 base_channels=-1,
Q
qingqing01 已提交
355 356 357
                 lr=1.0,
                 norm_type='bn',
                 norm_decay=0.,
F
Feng Ni 已提交
358 359
                 freeze_norm=True,
                 dcn_v2=False):
Q
qingqing01 已提交
360 361 362 363 364
        super(Blocks, self).__init__()

        self.blocks = []
        for i in range(count):
            conv_name = name_adapter.fix_layer_warp_name(stage_num, count, i)
365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400
            if depth >= 50:
                block = self.add_sublayer(
                    conv_name,
                    BottleNeck(
                        ch_in=ch_in if i == 0 else ch_out * 4,
                        ch_out=ch_out,
                        stride=2 if i == 0 and stage_num != 2 else 1,
                        shortcut=False if i == 0 else True,
                        name_adapter=name_adapter,
                        name=conv_name,
                        variant=variant,
                        groups=groups,
                        base_width=base_width,
                        base_channels=base_channels,
                        lr=lr,
                        norm_type=norm_type,
                        norm_decay=norm_decay,
                        freeze_norm=freeze_norm,
                        dcn_v2=dcn_v2))
            else:
                ch_in = ch_in // 4 if i > 0 else ch_in
                block = self.add_sublayer(
                    conv_name,
                    BasicBlock(
                        ch_in=ch_in if i == 0 else ch_out,
                        ch_out=ch_out,
                        stride=2 if i == 0 and stage_num != 2 else 1,
                        shortcut=False if i == 0 else True,
                        name_adapter=name_adapter,
                        name=conv_name,
                        variant=variant,
                        lr=lr,
                        norm_type=norm_type,
                        norm_decay=norm_decay,
                        freeze_norm=freeze_norm,
                        dcn_v2=dcn_v2))
Q
qingqing01 已提交
401 402 403 404 405 406 407 408 409 410 411 412
            self.blocks.append(block)

    def forward(self, inputs):
        block_out = inputs
        for block in self.blocks:
            block_out = block(block_out)
        return block_out


@register
@serializable
class ResNet(nn.Layer):
413 414
    __shared__ = ['norm_type']

Q
qingqing01 已提交
415 416 417
    def __init__(self,
                 depth=50,
                 variant='b',
418 419 420 421
                 lr_mult_list=[1.0, 1.0, 1.0, 1.0],
                 groups=1,
                 base_width=-1,
                 base_channels=-1,
Q
qingqing01 已提交
422 423 424 425 426
                 norm_type='bn',
                 norm_decay=0,
                 freeze_norm=True,
                 freeze_at=0,
                 return_idx=[0, 1, 2, 3],
F
Feng Ni 已提交
427
                 dcn_v2_stages=[-1],
Q
qingqing01 已提交
428 429
                 num_stages=4):
        super(ResNet, self).__init__()
430 431
        self._model_type = 'ResNet' if groups == 1 else 'ResNeXt'
        assert num_stages >= 1 and num_stages <= 4
Q
qingqing01 已提交
432 433 434 435 436 437 438 439 440 441 442 443 444 445
        self.depth = depth
        self.variant = variant
        self.norm_type = norm_type
        self.norm_decay = norm_decay
        self.freeze_norm = freeze_norm
        self.freeze_at = freeze_at
        if isinstance(return_idx, Integral):
            return_idx = [return_idx]
        assert max(return_idx) < num_stages, \
            'the maximum return index must smaller than num_stages, ' \
            'but received maximum return index is {} and num_stages ' \
            'is {}'.format(max(return_idx), num_stages)
        self.return_idx = return_idx
        self.num_stages = num_stages
446 447 448 449 450
        assert len(lr_mult_list) == 4, \
            "lr_mult_list length must be 4 but got {}".format(len(lr_mult_list))
        if isinstance(dcn_v2_stages, Integral):
            dcn_v2_stages = [dcn_v2_stages]
        assert max(dcn_v2_stages) < num_stages
Q
qingqing01 已提交
451

F
Feng Ni 已提交
452 453 454 455 456
        if isinstance(dcn_v2_stages, Integral):
            dcn_v2_stages = [dcn_v2_stages]
        assert max(dcn_v2_stages) < num_stages
        self.dcn_v2_stages = dcn_v2_stages

Q
qingqing01 已提交
457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478
        block_nums = ResNet_cfg[depth]
        na = NameAdapter(self)

        conv1_name = na.fix_c1_stage_name()
        if variant in ['c', 'd']:
            conv_def = [
                [3, 32, 3, 2, "conv1_1"],
                [32, 32, 3, 1, "conv1_2"],
                [32, 64, 3, 1, "conv1_3"],
            ]
        else:
            conv_def = [[3, 64, 7, 2, conv1_name]]
        self.conv1 = nn.Sequential()
        for (c_in, c_out, k, s, _name) in conv_def:
            self.conv1.add_sublayer(
                _name,
                ConvNormLayer(
                    ch_in=c_in,
                    ch_out=c_out,
                    filter_size=k,
                    stride=s,
                    name_adapter=na,
479
                    groups=1,
Q
qingqing01 已提交
480 481 482 483
                    act='relu',
                    norm_type=norm_type,
                    norm_decay=norm_decay,
                    freeze_norm=freeze_norm,
484
                    lr=1.0,
Q
qingqing01 已提交
485 486 487 488
                    name=_name))

        ch_in_list = [64, 256, 512, 1024]
        ch_out_list = [64, 128, 256, 512]
489 490 491 492
        self.expansion = 4 if depth >= 50 else 1

        self._out_channels = [self.expansion * v for v in ch_out_list]
        self._out_strides = [4, 8, 16, 32]
Q
qingqing01 已提交
493 494 495

        self.res_layers = []
        for i in range(num_stages):
496
            lr_mult = lr_mult_list[i]
Q
qingqing01 已提交
497 498 499 500 501
            stage_num = i + 2
            res_name = "res{}".format(stage_num)
            res_layer = self.add_sublayer(
                res_name,
                Blocks(
502 503 504
                    depth,
                    ch_in_list[i] // 4
                    if i > 0 and depth < 50 else ch_in_list[i],
Q
qingqing01 已提交
505 506 507 508
                    ch_out_list[i],
                    count=block_nums[i],
                    name_adapter=na,
                    stage_num=stage_num,
509 510 511 512
                    variant=variant,
                    groups=groups,
                    base_width=base_width,
                    base_channels=base_channels,
Q
qingqing01 已提交
513 514 515
                    lr=lr_mult,
                    norm_type=norm_type,
                    norm_decay=norm_decay,
F
Feng Ni 已提交
516 517
                    freeze_norm=freeze_norm,
                    dcn_v2=(i in self.dcn_v2_stages)))
Q
qingqing01 已提交
518 519
            self.res_layers.append(res_layer)

520 521 522 523 524 525 526 527
    @property
    def out_shape(self):
        return [
            ShapeSpec(
                channels=self._out_channels[i], stride=self._out_strides[i])
            for i in self.return_idx
        ]

Q
qingqing01 已提交
528 529 530
    def forward(self, inputs):
        x = inputs['image']
        conv1 = self.conv1(x)
531
        x = F.max_pool2d(conv1, kernel_size=3, stride=2, padding=1)
Q
qingqing01 已提交
532 533 534 535 536 537 538 539
        outs = []
        for idx, stage in enumerate(self.res_layers):
            x = stage(x)
            if idx == self.freeze_at:
                x.stop_gradient = True
            if idx in self.return_idx:
                outs.append(x)
        return outs
540 541 542 543


@register
class Res5Head(nn.Layer):
544
    def __init__(self, depth=50):
545
        super(Res5Head, self).__init__()
546 547 548
        feat_in, feat_out = [1024, 512]
        if depth < 50:
            feat_in = 256
549
        na = NameAdapter(self)
W
wangguanzhong 已提交
550 551
        self.res5 = Blocks(
            depth, feat_in, feat_out, count=3, name_adapter=na, stage_num=5)
552 553 554 555 556 557
        self.feat_out = feat_out if depth < 50 else feat_out * 4

    @property
    def out_shape(self):
        return [ShapeSpec(
            channels=self.feat_out,
558
            stride=16, )]
559 560 561 562

    def forward(self, roi_feat, stage=0):
        y = self.res5(roi_feat)
        return y