faster_rcnn.py 17.1 KB
Newer Older
1
2
3
4
5
6
7
8
9
from collections import OrderedDict

import torch
from torch import nn
import torch.nn.functional as F

from torchvision.ops import misc as misc_nn_ops
from torchvision.ops import MultiScaleRoIAlign

10
from ._utils import overwrite_eps
11
12
from ..utils import load_state_dict_from_url

13
from .anchor_utils import AnchorGenerator
14
from .generalized_rcnn import GeneralizedRCNN
15
from .rpn import RPNHead, RegionProposalNetwork
16
17
from .roi_heads import RoIHeads
from .transform import GeneralizedRCNNTransform
18
from .backbone_utils import resnet_fpn_backbone, _validate_resnet_trainable_layers
19
20
21
22
23
24
25
26


__all__ = [
    "FasterRCNN", "fasterrcnn_resnet50_fpn",
]


class FasterRCNN(GeneralizedRCNN):
27
28
29
30
31
32
33
34
    """
    Implements Faster R-CNN.

    The input to the model is expected to be a list of tensors, each of shape [C, H, W], one for each
    image, and should be in 0-1 range. Different images can have different sizes.

    The behavior of the model changes depending if it is in training or evaluation mode.

35
    During training, the model expects both the input tensors, as well as a targets (list of dictionary),
36
    containing:
37
38
        - boxes (FloatTensor[N, 4]): the ground-truth boxes in [x1, y1, x2, y2] format, with values of x
          between 0 and W and values of y between 0 and H
39
        - labels (Int64Tensor[N]): the class label for each ground-truth box
40

41
42
43
44
45
46
    The model returns a Dict[Tensor] during training, containing the classification and regression
    losses for both the RPN and the R-CNN.

    During inference, the model requires only the input tensors, and returns the post-processed
    predictions as a List[Dict[Tensor]], one for each input image. The fields of the Dict are as
    follows:
47
48
        - boxes (FloatTensor[N, 4]): the predicted boxes in [x1, y1, x2, y2] format, with values of x
          between 0 and W and values of y between 0 and H
49
        - labels (Int64Tensor[N]): the predicted labels for each image
50
        - scores (Tensor[N]): the scores or each prediction
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

    Arguments:
        backbone (nn.Module): the network used to compute the features for the model.
            It should contain a out_channels attribute, which indicates the number of output
            channels that each feature map has (and it should be the same for all feature maps).
            The backbone should return a single Tensor or and OrderedDict[Tensor].
        num_classes (int): number of output classes of the model (including the background).
            If box_predictor is specified, num_classes should be None.
        min_size (int): minimum size of the image to be rescaled before feeding it to the backbone
        max_size (int): maximum size of the image to be rescaled before feeding it to the backbone
        image_mean (Tuple[float, float, float]): mean values used for input normalization.
            They are generally the mean values of the dataset on which the backbone has been trained
            on
        image_std (Tuple[float, float, float]): std values used for input normalization.
            They are generally the std values of the dataset on which the backbone has been trained on
        rpn_anchor_generator (AnchorGenerator): module that generates the anchors for a set of feature
            maps.
        rpn_head (nn.Module): module that computes the objectness and regression deltas from the RPN
        rpn_pre_nms_top_n_train (int): number of proposals to keep before applying NMS during training
        rpn_pre_nms_top_n_test (int): number of proposals to keep before applying NMS during testing
        rpn_post_nms_top_n_train (int): number of proposals to keep after applying NMS during training
        rpn_post_nms_top_n_test (int): number of proposals to keep after applying NMS during testing
        rpn_nms_thresh (float): NMS threshold used for postprocessing the RPN proposals
        rpn_fg_iou_thresh (float): minimum IoU between the anchor and the GT box so that they can be
            considered as positive during training of the RPN.
        rpn_bg_iou_thresh (float): maximum IoU between the anchor and the GT box so that they can be
            considered as negative during training of the RPN.
        rpn_batch_size_per_image (int): number of anchors that are sampled during training of the RPN
            for computing the loss
        rpn_positive_fraction (float): proportion of positive anchors in a mini-batch during training
            of the RPN
        box_roi_pool (MultiScaleRoIAlign): the module which crops and resizes the feature maps in
            the locations indicated by the bounding boxes
        box_head (nn.Module): module that takes the cropped feature maps as input
        box_predictor (nn.Module): module that takes the output of box_head and returns the
            classification logits and box regression deltas.
        box_score_thresh (float): during inference, only return proposals with a classification score
            greater than box_score_thresh
        box_nms_thresh (float): NMS threshold for the prediction head. Used during inference
        box_detections_per_img (int): maximum number of detections per image, for all classes.
        box_fg_iou_thresh (float): minimum IoU between the proposals and the GT box so that they can be
            considered as positive during training of the classification head
        box_bg_iou_thresh (float): maximum IoU between the proposals and the GT box so that they can be
            considered as negative during training of the classification head
        box_batch_size_per_image (int): number of proposals that are sampled during training of the
            classification head
        box_positive_fraction (float): proportion of positive proposals in a mini-batch during training
            of the classification head
        bbox_reg_weights (Tuple[float, float, float, float]): weights for the encoding/decoding of the
            bounding boxes

    Example::

Gu-ni-kim's avatar
Gu-ni-kim committed
104
        >>> import torch
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
        >>> import torchvision
        >>> from torchvision.models.detection import FasterRCNN
        >>> from torchvision.models.detection.rpn import AnchorGenerator
        >>> # load a pre-trained model for classification and return
        >>> # only the features
        >>> backbone = torchvision.models.mobilenet_v2(pretrained=True).features
        >>> # FasterRCNN needs to know the number of
        >>> # output channels in a backbone. For mobilenet_v2, it's 1280
        >>> # so we need to add it here
        >>> backbone.out_channels = 1280
        >>>
        >>> # let's make the RPN generate 5 x 3 anchors per spatial
        >>> # location, with 5 different sizes and 3 different aspect
        >>> # ratios. We have a Tuple[Tuple[int]] because each feature
        >>> # map could potentially have different sizes and
        >>> # aspect ratios
        >>> anchor_generator = AnchorGenerator(sizes=((32, 64, 128, 256, 512),),
        >>>                                    aspect_ratios=((0.5, 1.0, 2.0),))
        >>>
        >>> # let's define what are the feature maps that we will
        >>> # use to perform the region of interest cropping, as well as
        >>> # the size of the crop after rescaling.
        >>> # if your backbone returns a Tensor, featmap_names is expected to
128
        >>> # be ['0']. More generally, the backbone should return an
129
130
        >>> # OrderedDict[Tensor], and in featmap_names you can choose which
        >>> # feature maps to use.
131
        >>> roi_pooler = torchvision.ops.MultiScaleRoIAlign(featmap_names=['0'],
132
133
134
135
136
137
138
139
        >>>                                                 output_size=7,
        >>>                                                 sampling_ratio=2)
        >>>
        >>> # put the pieces together inside a FasterRCNN model
        >>> model = FasterRCNN(backbone,
        >>>                    num_classes=2,
        >>>                    rpn_anchor_generator=anchor_generator,
        >>>                    box_roi_pool=roi_pooler)
140
141
142
143
144
        >>> model.eval()
        >>> x = [torch.rand(3, 300, 400), torch.rand(3, 500, 400)]
        >>> predictions = model(x)
    """

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
    def __init__(self, backbone, num_classes=None,
                 # transform parameters
                 min_size=800, max_size=1333,
                 image_mean=None, image_std=None,
                 # RPN parameters
                 rpn_anchor_generator=None, rpn_head=None,
                 rpn_pre_nms_top_n_train=2000, rpn_pre_nms_top_n_test=1000,
                 rpn_post_nms_top_n_train=2000, rpn_post_nms_top_n_test=1000,
                 rpn_nms_thresh=0.7,
                 rpn_fg_iou_thresh=0.7, rpn_bg_iou_thresh=0.3,
                 rpn_batch_size_per_image=256, rpn_positive_fraction=0.5,
                 # Box parameters
                 box_roi_pool=None, box_head=None, box_predictor=None,
                 box_score_thresh=0.05, box_nms_thresh=0.5, box_detections_per_img=100,
                 box_fg_iou_thresh=0.5, box_bg_iou_thresh=0.5,
                 box_batch_size_per_image=512, box_positive_fraction=0.25,
                 bbox_reg_weights=None):

        if not hasattr(backbone, "out_channels"):
            raise ValueError(
                "backbone should contain an attribute out_channels "
                "specifying the number of output channels (assumed to be the "
                "same for all the levels)")

        assert isinstance(rpn_anchor_generator, (AnchorGenerator, type(None)))
        assert isinstance(box_roi_pool, (MultiScaleRoIAlign, type(None)))

        if num_classes is not None:
            if box_predictor is not None:
                raise ValueError("num_classes should be None when box_predictor is specified")
        else:
            if box_predictor is None:
                raise ValueError("num_classes should not be None when box_predictor "
                                 "is not specified")

        out_channels = backbone.out_channels

        if rpn_anchor_generator is None:
            anchor_sizes = ((32,), (64,), (128,), (256,), (512,))
            aspect_ratios = ((0.5, 1.0, 2.0),) * len(anchor_sizes)
            rpn_anchor_generator = AnchorGenerator(
                anchor_sizes, aspect_ratios
            )
        if rpn_head is None:
            rpn_head = RPNHead(
                out_channels, rpn_anchor_generator.num_anchors_per_location()[0]
            )

        rpn_pre_nms_top_n = dict(training=rpn_pre_nms_top_n_train, testing=rpn_pre_nms_top_n_test)
        rpn_post_nms_top_n = dict(training=rpn_post_nms_top_n_train, testing=rpn_post_nms_top_n_test)

        rpn = RegionProposalNetwork(
            rpn_anchor_generator, rpn_head,
            rpn_fg_iou_thresh, rpn_bg_iou_thresh,
            rpn_batch_size_per_image, rpn_positive_fraction,
            rpn_pre_nms_top_n, rpn_post_nms_top_n, rpn_nms_thresh)

        if box_roi_pool is None:
            box_roi_pool = MultiScaleRoIAlign(
eellison's avatar
eellison committed
204
                featmap_names=['0', '1', '2', '3'],
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
                output_size=7,
                sampling_ratio=2)

        if box_head is None:
            resolution = box_roi_pool.output_size[0]
            representation_size = 1024
            box_head = TwoMLPHead(
                out_channels * resolution ** 2,
                representation_size)

        if box_predictor is None:
            representation_size = 1024
            box_predictor = FastRCNNPredictor(
                representation_size,
                num_classes)

        roi_heads = RoIHeads(
            # Box
            box_roi_pool, box_head, box_predictor,
            box_fg_iou_thresh, box_bg_iou_thresh,
            box_batch_size_per_image, box_positive_fraction,
            bbox_reg_weights,
            box_score_thresh, box_nms_thresh, box_detections_per_img)

        if image_mean is None:
            image_mean = [0.485, 0.456, 0.406]
        if image_std is None:
            image_std = [0.229, 0.224, 0.225]
        transform = GeneralizedRCNNTransform(min_size, max_size, image_mean, image_std)

        super(FasterRCNN, self).__init__(backbone, rpn, roi_heads, transform)


class TwoMLPHead(nn.Module):
    """
240
241
242
243
244
    Standard heads for FPN-based models

    Arguments:
        in_channels (int): number of input channels
        representation_size (int): size of the intermediate representation
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
    """

    def __init__(self, in_channels, representation_size):
        super(TwoMLPHead, self).__init__()

        self.fc6 = nn.Linear(in_channels, representation_size)
        self.fc7 = nn.Linear(representation_size, representation_size)

    def forward(self, x):
        x = x.flatten(start_dim=1)

        x = F.relu(self.fc6(x))
        x = F.relu(self.fc7(x))

        return x


class FastRCNNPredictor(nn.Module):
263
264
265
266
267
268
269
270
271
    """
    Standard classification + bounding box regression layers
    for Fast R-CNN.

    Arguments:
        in_channels (int): number of input channels
        num_classes (int): number of output classes (including background)
    """

272
273
274
275
276
277
    def __init__(self, in_channels, num_classes):
        super(FastRCNNPredictor, self).__init__()
        self.cls_score = nn.Linear(in_channels, num_classes)
        self.bbox_pred = nn.Linear(in_channels, num_classes * 4)

    def forward(self, x):
eellison's avatar
eellison committed
278
        if x.dim() == 4:
279
280
281
282
283
284
285
286
            assert list(x.shape[2:]) == [1, 1]
        x = x.flatten(start_dim=1)
        scores = self.cls_score(x)
        bbox_deltas = self.bbox_pred(x)

        return scores, bbox_deltas


287
288
289
290
291
292
293
model_urls = {
    'fasterrcnn_resnet50_fpn_coco':
        'https://download.pytorch.org/models/fasterrcnn_resnet50_fpn_coco-258fb6c6.pth',
}


def fasterrcnn_resnet50_fpn(pretrained=False, progress=True,
294
                            num_classes=91, pretrained_backbone=True, trainable_backbone_layers=None, **kwargs):
295
296
297
    """
    Constructs a Faster R-CNN model with a ResNet-50-FPN backbone.

298
299
300
301
302
    The input to the model is expected to be a list of tensors, each of shape ``[C, H, W]``, one for each
    image, and should be in ``0-1`` range. Different images can have different sizes.

    The behavior of the model changes depending if it is in training or evaluation mode.

303
    During training, the model expects both the input tensors, as well as a targets (list of dictionary),
304
    containing:
305
306
        - boxes (``FloatTensor[N, 4]``): the ground-truth boxes in ``[x1, y1, x2, y2]`` format, with values of ``x``
          between ``0`` and ``W`` and values of ``y`` between ``0`` and ``H``
307
        - labels (``Int64Tensor[N]``): the class label for each ground-truth box
308
309
310
311
312
313
314

    The model returns a ``Dict[Tensor]`` during training, containing the classification and regression
    losses for both the RPN and the R-CNN.

    During inference, the model requires only the input tensors, and returns the post-processed
    predictions as a ``List[Dict[Tensor]]``, one for each input image. The fields of the ``Dict`` are as
    follows:
315
316
        - boxes (``FloatTensor[N, 4]``): the predicted boxes in ``[x1, y1, x2, y2]`` format, with values of ``x``
          between ``0`` and ``W`` and values of ``y`` between ``0`` and ``H``
317
        - labels (``Int64Tensor[N]``): the predicted labels for each image
318
319
        - scores (``Tensor[N]``): the scores or each prediction

320
321
    Faster R-CNN is exportable to ONNX for a fixed batch size with inputs images of fixed size.

322
323
324
    Example::

        >>> model = torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained=True)
325
326
327
        >>> # For training
        >>> images, boxes = torch.rand(4, 3, 600, 1200), torch.rand(4, 11, 4)
        >>> labels = torch.randint(1, 91, (4, 11))
328
        >>> images = list(image for image in images)
329
        >>> targets = []
330
331
332
        >>> for i in range(len(images)):
        >>>     d = {}
        >>>     d['boxes'] = boxes[i]
333
        >>>     d['labels'] = labels[i]
334
        >>>     targets.append(d)
335
336
337
        >>> output = model(images, targets)
        >>> # For inference
        >>> model.eval()
338
339
        >>> x = [torch.rand(3, 300, 400), torch.rand(3, 500, 400)]
        >>> predictions = model(x)
340
341
342
        >>>
        >>> # optionally, if you want to export the model to ONNX:
        >>> torch.onnx.export(model, x, "faster_rcnn.onnx", opset_version = 11)
343

344
345
346
    Arguments:
        pretrained (bool): If True, returns a model pre-trained on COCO train2017
        progress (bool): If True, displays a progress bar of the download to stderr
347
348
349
350
        pretrained_backbone (bool): If True, returns a model with backbone pre-trained on Imagenet
        num_classes (int): number of output classes of the model (including the background)
        trainable_backbone_layers (int): number of trainable (not frozen) resnet layers starting from final block.
            Valid values are between 0 and 5, with 5 meaning all backbone layers are trainable.
351
    """
352
353
354
355
    # check default parameters and by default set it to 3 if possible
    trainable_backbone_layers = _validate_resnet_trainable_layers(
        pretrained or pretrained_backbone, trainable_backbone_layers)

356
357
358
    if pretrained:
        # no need to download the backbone if pretrained is set
        pretrained_backbone = False
359
    backbone = resnet_fpn_backbone('resnet50', pretrained_backbone, trainable_layers=trainable_backbone_layers)
360
361
    model = FasterRCNN(backbone, num_classes, **kwargs)
    if pretrained:
362
363
364
        state_dict = load_state_dict_from_url(model_urls['fasterrcnn_resnet50_fpn_coco'],
                                              progress=progress)
        model.load_state_dict(state_dict)
365
        overwrite_eps(model, 0.0)
366
    return model