# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() NECKS = Registry() ONE_STAGE_HEADS = Registry() PANOPTIC_HEADS = Registry() RPN_HEADS = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() ROI_BOX_PREDICTOR = Registry() ROI_KEYPOINT_FEATURE_EXTRACTORS = Registry() ROI_KEYPOINT_PREDICTOR = Registry() ROI_MASK_FEATURE_EXTRACTORS = Registry() ROI_MASK_PREDICTOR = Registry()
num_groups=num_groups,
stride_in_1x1=stride_in_1x1,
stride=stride,
dilation=dilation,
norm_func=group_norm)
class StemWithGN(BaseStem):
def __init__(self, cfg):
super(StemWithGN, self).__init__(cfg, norm_func=group_norm)
_TRANSFORMATION_MODULES = Registry({
"BottleneckWithFixedBatchNorm":
BottleneckWithFixedBatchNorm,
"BottleneckWithGN":
BottleneckWithGN,
"BottleneckFixedBN_Mish":
BottleneckFixedBN_Mish
})
_STEM_MODULES = Registry({
"StemWithFixedBatchNorm": StemWithFixedBatchNorm,
"RootStemWithFixedBatchNorm": RootStemWithFixedBatchNorm,
"StemWithGN": StemWithGN,
"RootStemWithGN": RootStemWithGN,
})
_STAGE_SPECS = Registry({
"R-50-C4": ResNet50StagesTo4,
"R-50-C5": ResNet50StagesTo5,
"R-101-C4": ResNet101StagesTo4,
import sys
import torch
import torch.nn as nn
from torch import Tensor
from torch.nn import Module
from maskrcnn_benchmark.utils.registry import Registry
sys.setrecursionlimit(10000)
MODULE_FLOPS_COMP = Registry()
@MODULE_FLOPS_COMP.register("Conv2d")
def build_conv2d_flops(module: Module, oup: Tensor):
kh, kw = module.kernel_size
in_channels = module.in_channels
out_channels = module.out_channels
groups = module.groups
h, w = oup.shape[-2:]
return in_channels * out_channels * w * h * kh * kw // groups
@MODULE_FLOPS_COMP.register("Linear")
def build_linear_flops(module: Module, oup: Tensor):
in_features = module.in_features
out_features = module.out_features
return in_features * out_features
'eSE': True
}
VoVNet99_eSE_FPNStagesTo5 = {
'config_stage_ch': [128, 160, 192, 224],
'config_concat_ch': [256, 512, 768, 1024],
'layer_per_block': 5,
'block_per_stage': [1, 3, 9, 3],
'eSE': True
}
_STAGE_SPECS = Registry({
"V-19-eSE-FPN-RETINANET": VoVNet19_eSE_FPNStagesTo5,
"V-39-eSE-FPN-RETINANET": VoVNet39_eSE_FPNStagesTo5,
"V-57-eSE-FPN-RETINANET": VoVNet57_eSE_FPNStagesTo5,
"V-99-eSE-FPN-RETINANET": VoVNet99_eSE_FPNStagesTo5,
"V-19-eSE-FPN": VoVNet19_eSE_FPNStagesTo5,
"V-39-eSE-FPN": VoVNet39_eSE_FPNStagesTo5,
"V-57-eSE-FPN": VoVNet57_eSE_FPNStagesTo5,
"V-99-eSE-FPN": VoVNet99_eSE_FPNStagesTo5
})
def freeze_bn_params(m):
"""Freeze all the weights by setting requires_grad to False
"""
m.eval()
for p in m.parameters():
p.requires_grad = False
def conv3x3(in_channels,
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() RPN_HEADS = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() ROI_BOX_PREDICTOR = Registry() ROI_KEYPOINT_FEATURE_EXTRACTORS = Registry() ROI_KEYPOINT_PREDICTOR = Registry() ROI_MASK_FEATURE_EXTRACTORS = Registry() ROI_MASK_PREDICTOR = Registry() CLASSIFIER_ENCODER = Registry() CLASSIFIER_DECODER = Registry()
from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() RPN_HEADS = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() ROI_BOX_PREDICTOR = Registry() ROI_KEYPOINT_FEATURE_EXTRACTORS = Registry() ROI_KEYPOINT_PREDICTOR = Registry() ROI_MASK_FEATURE_EXTRACTORS = Registry() ROI_MASK_PREDICTOR = Registry()
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() ROI_BOX_PREDICTOR = Registry() RPN_HEADS = Registry()
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() RPN_HEADS = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() ROI_BOX_PREDICTOR = Registry() ROI_KEYPOINT_FEATURE_EXTRACTORS = Registry() ROI_KEYPOINT_PREDICTOR = Registry() ROI_MASK_FEATURE_EXTRACTORS = Registry() ROI_MASK_PREDICTOR = Registry() SEG_BRANCH = Registry() INTER_MODULE = Registry()
# Copyright (c) 2021 Microsoft Corporation. Licensed under the MIT license. from maskrcnn_benchmark.utils.registry import Registry ROI_ATTRIBUTE_FEATURE_EXTRACTORS = Registry() ROI_ATTRIBUTE_PREDICTOR = Registry() ROI_RELATION_FEATURE_EXTRACTORS = Registry() ROI_RELATION_PREDICTOR = Registry() ROI_RELATION_BOX_FEATURE_EXTRACTORS = Registry() ROI_RELATION_BOX_PREDICTOR = Registry()
from maskrcnn_benchmark.utils.registry import Registry SIAMESE_TRACKER = Registry() TRACKER_SAMPLER = Registry()
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() RPN_HEADS = Registry() SEMANTIC_HEADS = Registry() UPSAMPLE_MODULES = Registry()
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() RPN_HEADS = Registry() STEREO_RPN_V1_HEADS = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() ROI_BOX_PREDICTOR = Registry() ROI_KEYPOINT_FEATURE_EXTRACTORS = Registry() ROI_KEYPOINT_PREDICTOR = Registry() ROI_MASK_FEATURE_EXTRACTORS = Registry() ROI_MASK_PREDICTOR = Registry() ROI_BOX_STEREO_V1_FEATURE_EXTRACTORS = Registry() ROI_BOX_STEREO_V1_PREDICTOR = Registry() ROI_BOX_STEREO_V2_FEATURE_EXTRACTORS = Registry() ROI_BOX_STEREO_V2_PREDICTOR = Registry() ROI_BOX_TD_V1_FEATURE_EXTRACTORS = Registry() ROI_BOX_TD_V1_PREDICTOR = Registry() ROI_BOX_TD_V2_FEATURE_EXTRACTORS = Registry() ROI_BOX_TD_V2_PREDICTOR = Registry() ROI_BOX_TD_V3_FEATURE_EXTRACTORS = Registry()
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() RPN_HEADS = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() ROI_BOX_PREDICTOR = Registry() ROI_KEYPOINT_FEATURE_EXTRACTORS = Registry() ROI_KEYPOINT_PREDICTOR = Registry() ROI_MASK_FEATURE_EXTRACTORS = Registry() ROI_MASK_PREDICTOR = Registry() ROI_PRED_FEATURE_EXTRACTORS = Registry() ROI_PRED_PREDICTOR = Registry() DECODER_CLS_FEATURE_EXTRACTORS = Registry() DECODER_CLS_PREDICTOR = Registry() ATTENTION_MECHANISM = Registry() POOLER = Registry()
"""darknet19 from https://pjreddie.com/darknet/imagenet/"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals
import torch
import torch.nn as nn
import torch.nn.functional as F
from maskrcnn_benchmark.layers import FrozenBatchNorm2d
from maskrcnn_benchmark.utils.registry import Registry
_NORMALIZATION_MODULES = Registry({
"FrozenBatchNorm2d": FrozenBatchNorm2d,
"BatchNorm2d": nn.BatchNorm2d,
})
def add_stage(
n,
dim_in,
dim_out1,
dim_inner,
normalization
):
"""Add a ResNet stage to the model by stacking n residual blocks."""
blocks = []
for i in range(n):
blocks.append(Xception_block(
dim_in,
dim_out1,
else:
data = pickle.load(f)
if "blobs" in data:
weights = data["blobs"]
else:
weights = data
return weights
_C2_STAGE_NAMES = {
"R-50": ["1.2", "2.3", "3.5", "4.2"],
"R-101": ["1.2", "2.3", "3.22", "4.2"],
"R-152": ["1.2", "2.7", "3.35", "4.2"],
}
C2_FORMAT_LOADER = Registry()
@C2_FORMAT_LOADER.register("R-50-C4")
@C2_FORMAT_LOADER.register("R-50-C5")
@C2_FORMAT_LOADER.register("R-101-C4")
@C2_FORMAT_LOADER.register("R-101-C5")
@C2_FORMAT_LOADER.register("R-50-FPN")
@C2_FORMAT_LOADER.register("R-50-FPN-RETINANET")
@C2_FORMAT_LOADER.register("R-101-FPN")
@C2_FORMAT_LOADER.register("R-101-FPN-RETINANET")
@C2_FORMAT_LOADER.register("R-152-FPN")
def load_resnet_c2_format(cfg, f):
state_dict = _load_c2_pickled_weights(f)
conv_body = cfg.MODEL.BACKBONE.CONV_BODY
arch = conv_body.replace("-C4", "").replace("-C5", "").replace("-FPN", "")
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. from maskrcnn_benchmark.utils.registry import Registry BACKBONES = Registry() RPN_HEADS = Registry() ROI_BOX_FEATURE_EXTRACTORS = Registry() ROI_BOX_PREDICTOR = Registry() ROI_ATTRIBUTE_FEATURE_EXTRACTORS = Registry() ROI_ATTRIBUTE_PREDICTOR = Registry() ROI_KEYPOINT_FEATURE_EXTRACTORS = Registry() ROI_KEYPOINT_PREDICTOR = Registry() ROI_MASK_FEATURE_EXTRACTORS = Registry() ROI_MASK_PREDICTOR = Registry() ROI_RELATION_FEATURE_EXTRACTORS = Registry() ROI_RELATION_PREDICTOR = Registry()
def dla_169(dcn_config, feature_only=True, batch_norm=FrozenBatchNorm2d):
model = DLA([1, 1, 2, 3, 5, 1], [16, 32, 128, 256, 512, 1024],
block=DlaBottleneck,
residual_root=True,
batch_norm=batch_norm,
feature_only=feature_only,
dcn_config=dcn_config)
return model
BACKBONE = Registry({
"DLA-34-FPN": dla_34,
"DLA-46-C-FPN": dla_46_c,
"DLA-46-XC-FPN": dla_46_xc,
"DLA-60-FPN": dla_60,
"DLA-60-RES2NET-FPN": dla60_res2net,
"DLA-102-FPN": dla_102,
"DLA-169-FPN": dla_169
})
BACKBONE_ARCH = {
"DLA-34-FPN": "dla34",
"DLA-46-C-FPN": "dla_46_c",
"DLA-46-XC-FPN": "dla_46_xc",
"DLA-60-FPN": "dla_60",
"DLA-60-RES2NET-FPN": "dla60_res2net",
"DLA-102-FPN": "dla_102",
"DLA-169-FPN": "dla_169"
}
):
super(TridentBottleneckWithFixedBatchNorm, self).__init__(
in_channels=in_channels,
bottleneck_channels=bottleneck_channels,
out_channels=out_channels,
num_groups=num_groups,
stride_in_1x1=stride_in_1x1,
stride=stride,
dilation=dilation,
norm_func=TridentFrozenBatchNorm2d,
)
_TRANSFORMATION_MODULES = Registry({
"BottleneckWithFixedBatchNorm": BottleneckWithFixedBatchNorm,
"BottleneckWithGN": BottleneckWithGN,
"TridentBottleneckWithFixedBatchNorm": TridentBottleneckWithFixedBatchNorm,
})
_STEM_MODULES = Registry({
"StemWithFixedBatchNorm": StemWithFixedBatchNorm,
"StemWithGN": StemWithGN,
})
_STAGE_SPECS = Registry({
"T-R-50-C5": TridentResNet50StagesTo5,
"T-R-101-C5": TridentResNet101StagesTo5,
"T-R-50-FPN": TridentResNet50FPNStagesTo5,
"T-R-101-FPN": TridentResNet101FPNStagesTo5,
"T-R-152-FPN": TridentResNet152FPNStagesTo5,
})
def __init__(self, cfg):
super(StemWithFixedBatchNorm, self).__init__()
out_channels = cfg.MODEL.SENET.STEM_OUT_CHANNELS
self.conv1 = Conv2d(
3, out_channels, kernel_size=7, stride=2, padding=3, bias=False
)
self.bn1 = FrozenBatchNorm2d(out_channels)
def forward(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = F.relu_(x)
x = F.max_pool2d(x, kernel_size=3, stride=2, padding=1)
return x
_TRANSFORMATION_MODULES = Registry({
"BottleneckWithFixedBatchNorm": BottleneckWithFixedBatchNorm
})
_STEM_MODULES = Registry({"StemWithFixedBatchNorm": StemWithFixedBatchNorm})
_STAGE_SPECS = Registry({
"SE-50-C4": SENet50StagesTo4,
"SE-50-C5": SENet50StagesTo5,
"SE-50-FPN": SENet50FPNStagesTo5,
"SE-101-FPN": SENet101FPNStagesTo5,
})
from maskrcnn_benchmark.utils.registry import Registry ROI_GRAPH_FEATURE_EXTRACTORS = Registry() ROI_GRAPH_PREDICTOR = Registry()
num_groups=num_groups,
stride_in_1x1=stride_in_1x1,
stride=stride,
dilation=dilation,
norm_func=group_norm,
dcn_config=dcn_config)
class StemWithGN(BaseStem):
def __init__(self, cfg):
super(StemWithGN, self).__init__(cfg, norm_func=group_norm)
_TRANSFORMATION_MODULES = Registry({
"BottleneckWithFixedBatchNorm": BottleneckWithFixedBatchNorm,
"BottleneckWithBatchNorm": BottleneckWithBatchNorm,
"BottleneckWithGN": BottleneckWithGN,
})
_STEM_MODULES = Registry({
"StemWithFixedBatchNorm": StemWithFixedBatchNorm,
"StemWithBatchNorm": StemWithBatchNorm,
"StemWithGN": StemWithGN,
})
_STAGE_SPECS = Registry({
"R-50-C4": ResNet50StagesTo4,
"R-50-C5": ResNet50StagesTo5,
"R-101-C4": ResNet101StagesTo4,
"R-101-C5": ResNet101StagesTo5,
"R-50-FPN": ResNet50FPNStagesTo5,
super(BottleneckWithGN,
self).__init__(in_channels=in_channels,
bottleneck_channels=bottleneck_channels,
out_channels=out_channels,
num_groups=num_groups,
stride_in_1x1=stride_in_1x1,
stride=stride,
dilation=dilation,
norm_func=group_norm)
class StemWithGN(BaseStem):
def __init__(self, cfg):
super(StemWithGN, self).__init__(cfg, norm_func=group_norm)
_TRANSFORMATION_MODULES = Registry({
"BottleneckWithFixedBatchNorm": BottleneckWithFixedBatchNorm,
"BottleneckWithGN": BottleneckWithGN,
})
_STEM_MODULES = Registry({
"StemWithFixedBatchNorm": StemWithFixedBatchNorm,
"StemWithGN": StemWithGN,
})
_STAGE_SPECS = Registry({
"OR-50-FPN": OR_ResNet50FPNStagesTo5,
"OR-101-FPN": OR_ResNet101FPNStagesTo5,
})
'block_per_stage': [1, 2, 6, 3]
}
VoVNet93FPNStagesTo5 = {
'config_stage_ch': [128, 160, 192, 224],
'config_concat_ch': [256, 512, 768, 1024],
'layer_per_block': 5,
'block_per_stage': [1, 3, 8, 3]
}
_STAGE_SPECS = Registry({
"V-27-FPN": VoVNet27FPNStagesTo5,
"V-39-FPN": VoVNet39FPNStagesTo5,
"V-57-FPN": VoVNet57FPNStagesTo5,
"V-75-FPN": VoVNet75FPNStagesTo5,
"V-93-FPN": VoVNet93FPNStagesTo5,
"V-27-FPN-RETINANET": VoVNet27FPNStagesTo5,
"V-39-FPN-RETINANET": VoVNet39FPNStagesTo5,
"V-57-FPN-RETINANET": VoVNet57FPNStagesTo5,
"V-75-FPN-RETINANET": VoVNet75FPNStagesTo5,
"V-93-FPN-RETINANET": VoVNet93FPNStagesTo5
})
def freeze_bn_params(m):
"""Freeze all the weights by setting requires_grad to False
"""
m.eval()
for p in m.parameters():
p.requires_grad = False
def conv3x3(in_channels, out_channels, module_name, postfix, stride=1, groups=1, kernel_size=3, padding=1):
"""3x3 convolution with padding"""
bottleneck_channels,
out_channels,
num_groups,
stride_in_1x1,
stride,
dilation=dilation,
norm_func=FrozenBatchNorm2d,
deform=deform,
)
)
stride = 1
in_channels = out_channels
return nn.Sequential(*blocks)
_TRANSFORMATION_MODULES = Registry({
"BottleneckWithFixedBatchNorm": BottleneckWithFixedBatchNorm,
"BottleneckWithGN": BottleneckWithGN,
})
_STEM_MODULES = Registry({
"StemWithFixedBatchNorm": StemWithFixedBatchNorm,
"StemWithGN": StemWithGN,
})
_STAGE_SPECS = Registry({
"D-59-C6": DetNet59StagesTo6,
"D-59-FPN": DetNet59FPNStagesTo6,
"D-59-D-FPN": DetNet59DeformFPNStagesTo6,
})