File size: 7,972 Bytes
3f75218
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# ------------------------------------------------------------------------
# Grounding DINO
# url: https://github.com/IDEA-Research/GroundingDINO
# Copyright (c) 2023 IDEA. All Rights Reserved.
# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
# ------------------------------------------------------------------------
# Conditional DETR
# Copyright (c) 2021 Microsoft. All Rights Reserved.
# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
# ------------------------------------------------------------------------
# Copied from DETR (https://github.com/facebookresearch/detr)
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
# ------------------------------------------------------------------------

"""
Backbone modules.
"""

from typing import Dict, List

import torch
import torch.nn.functional as F
import torchvision
from torch import nn
from torchvision.models._utils import IntermediateLayerGetter

from groundingdino.util.misc import NestedTensor, clean_state_dict, is_main_process

from .position_encoding import build_position_encoding
from .swin_transformer import build_swin_transformer


class FrozenBatchNorm2d(torch.nn.Module):
    """
    BatchNorm2d where the batch statistics and the affine parameters are fixed.

    Copy-paste from torchvision.misc.ops with added eps before rqsrt,
    without which any other models than torchvision.models.resnet[18,34,50,101]
    produce nans.
    """

    def __init__(self, n):
        super(FrozenBatchNorm2d, self).__init__()
        self.register_buffer("weight", torch.ones(n))
        self.register_buffer("bias", torch.zeros(n))
        self.register_buffer("running_mean", torch.zeros(n))
        self.register_buffer("running_var", torch.ones(n))

    def _load_from_state_dict(
        self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs
    ):
        num_batches_tracked_key = prefix + "num_batches_tracked"
        if num_batches_tracked_key in state_dict:
            del state_dict[num_batches_tracked_key]

        super(FrozenBatchNorm2d, self)._load_from_state_dict(
            state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs
        )

    def forward(self, x):
        # move reshapes to the beginning
        # to make it fuser-friendly
        w = self.weight.reshape(1, -1, 1, 1)
        b = self.bias.reshape(1, -1, 1, 1)
        rv = self.running_var.reshape(1, -1, 1, 1)
        rm = self.running_mean.reshape(1, -1, 1, 1)
        eps = 1e-5
        scale = w * (rv + eps).rsqrt()
        bias = b - rm * scale
        return x * scale + bias


class BackboneBase(nn.Module):
    def __init__(
        self,
        backbone: nn.Module,
        train_backbone: bool,
        num_channels: int,
        return_interm_indices: list,
    ):
        super().__init__()
        for name, parameter in backbone.named_parameters():
            if (
                not train_backbone
                or "layer2" not in name
                and "layer3" not in name
                and "layer4" not in name
            ):
                parameter.requires_grad_(False)

        return_layers = {}
        for idx, layer_index in enumerate(return_interm_indices):
            return_layers.update(
                {"layer{}".format(5 - len(return_interm_indices) + idx): "{}".format(layer_index)}
            )

        # if len:
        #     if use_stage1_feature:
        #         return_layers = {"layer1": "0", "layer2": "1", "layer3": "2", "layer4": "3"}
        #     else:
        #         return_layers = {"layer2": "0", "layer3": "1", "layer4": "2"}
        # else:
        #     return_layers = {'layer4': "0"}
        self.body = IntermediateLayerGetter(backbone, return_layers=return_layers)
        self.num_channels = num_channels

    def forward(self, tensor_list: NestedTensor):
        xs = self.body(tensor_list.tensors)
        out: Dict[str, NestedTensor] = {}
        for name, x in xs.items():
            m = tensor_list.mask
            assert m is not None
            mask = F.interpolate(m[None].float(), size=x.shape[-2:]).to(torch.bool)[0]
            out[name] = NestedTensor(x, mask)
        # import ipdb; ipdb.set_trace()
        return out


class Backbone(BackboneBase):
    """ResNet backbone with frozen BatchNorm."""

    def __init__(
        self,
        name: str,
        train_backbone: bool,
        dilation: bool,
        return_interm_indices: list,
        batch_norm=FrozenBatchNorm2d,
    ):
        if name in ["resnet18", "resnet34", "resnet50", "resnet101"]:
            backbone = getattr(torchvision.models, name)(
                replace_stride_with_dilation=[False, False, dilation],
                pretrained=is_main_process(),
                norm_layer=batch_norm,
            )
        else:
            raise NotImplementedError("Why you can get here with name {}".format(name))
        # num_channels = 512 if name in ('resnet18', 'resnet34') else 2048
        assert name not in ("resnet18", "resnet34"), "Only resnet50 and resnet101 are available."
        assert return_interm_indices in [[0, 1, 2, 3], [1, 2, 3], [3]]
        num_channels_all = [256, 512, 1024, 2048]
        num_channels = num_channels_all[4 - len(return_interm_indices) :]
        super().__init__(backbone, train_backbone, num_channels, return_interm_indices)


class Joiner(nn.Sequential):
    def __init__(self, backbone, position_embedding):
        super().__init__(backbone, position_embedding)

    def forward(self, tensor_list: NestedTensor):
        xs = self[0](tensor_list)
        out: List[NestedTensor] = []
        pos = []
        for name, x in xs.items():
            out.append(x)
            # position encoding
            pos.append(self[1](x).to(x.tensors.dtype))

        return out, pos


def build_backbone(args):
    """
    Useful args:
        - backbone: backbone name
        - lr_backbone:
        - dilation
        - return_interm_indices: available: [0,1,2,3], [1,2,3], [3]
        - backbone_freeze_keywords:
        - use_checkpoint: for swin only for now

    """
    position_embedding = build_position_encoding(args)
    train_backbone = True
    if not train_backbone:
        raise ValueError("Please set lr_backbone > 0")
    return_interm_indices = args.return_interm_indices
    assert return_interm_indices in [[0, 1, 2, 3], [1, 2, 3], [3]]
    args.backbone_freeze_keywords
    use_checkpoint = getattr(args, "use_checkpoint", False)

    if args.backbone in ["resnet50", "resnet101"]:
        backbone = Backbone(
            args.backbone,
            train_backbone,
            args.dilation,
            return_interm_indices,
            batch_norm=FrozenBatchNorm2d,
        )
        bb_num_channels = backbone.num_channels
    elif args.backbone in [
        "swin_T_224_1k",
        "swin_B_224_22k",
        "swin_B_384_22k",
        "swin_L_224_22k",
        "swin_L_384_22k",
    ]:
        pretrain_img_size = int(args.backbone.split("_")[-2])
        backbone = build_swin_transformer(
            args.backbone,
            pretrain_img_size=pretrain_img_size,
            out_indices=tuple(return_interm_indices),
            dilation=False,
            use_checkpoint=use_checkpoint,
        )

        bb_num_channels = backbone.num_features[4 - len(return_interm_indices) :]
    else:
        raise NotImplementedError("Unknown backbone {}".format(args.backbone))

    assert len(bb_num_channels) == len(
        return_interm_indices
    ), f"len(bb_num_channels) {len(bb_num_channels)} != len(return_interm_indices) {len(return_interm_indices)}"

    model = Joiner(backbone, position_embedding)
    model.num_channels = bb_num_channels
    assert isinstance(
        bb_num_channels, List
    ), "bb_num_channels is expected to be a List but {}".format(type(bb_num_channels))
    # import ipdb; ipdb.set_trace()
    return model