File size: 11,930 Bytes
61c2d32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) Facebook, Inc. and its affiliates.
import numpy as np
import fvcore.nn.weight_init as weight_init
import torch
from torch import nn
from torch.nn import functional as F

from detectron2.layers import ShapeSpec, cat
from detectron2.utils.events import get_event_storage
from detectron2.utils.registry import Registry

POINT_HEAD_REGISTRY = Registry("POINT_HEAD")
POINT_HEAD_REGISTRY.__doc__ = """
Registry for point heads, which makes prediction for a given set of per-point features.

The registered object will be called with `obj(cfg, input_shape)`.
"""


def roi_mask_point_loss(mask_logits, instances, point_labels):
    """
    Compute the point-based loss for instance segmentation mask predictions
    given point-wise mask prediction and its corresponding point-wise labels.
    Args:
        mask_logits (Tensor): A tensor of shape (R, C, P) or (R, 1, P) for class-specific or
            class-agnostic, where R is the total number of predicted masks in all images, C is the
            number of foreground classes, and P is the number of points sampled for each mask.
            The values are logits.
        instances (list[Instances]): A list of N Instances, where N is the number of images
            in the batch. These instances are in 1:1 correspondence with the `mask_logits`. So, i_th
            elememt of the list contains R_i objects and R_1 + ... + R_N is equal to R.
            The ground-truth labels (class, box, mask, ...) associated with each instance are stored
            in fields.
        point_labels (Tensor): A tensor of shape (R, P), where R is the total number of
            predicted masks and P is the number of points for each mask.
            Labels with value of -1 will be ignored.
    Returns:
        point_loss (Tensor): A scalar tensor containing the loss.
    """
    with torch.no_grad():
        cls_agnostic_mask = mask_logits.size(1) == 1
        total_num_masks = mask_logits.size(0)

        gt_classes = []
        for instances_per_image in instances:
            if len(instances_per_image) == 0:
                continue

            if not cls_agnostic_mask:
                gt_classes_per_image = instances_per_image.gt_classes.to(dtype=torch.int64)
                gt_classes.append(gt_classes_per_image)

    gt_mask_logits = point_labels
    point_ignores = point_labels == -1
    if gt_mask_logits.shape[0] == 0:
        return mask_logits.sum() * 0

    assert gt_mask_logits.numel() > 0, gt_mask_logits.shape

    if cls_agnostic_mask:
        mask_logits = mask_logits[:, 0]
    else:
        indices = torch.arange(total_num_masks)
        gt_classes = cat(gt_classes, dim=0)
        mask_logits = mask_logits[indices, gt_classes]

    # Log the training accuracy (using gt classes and 0.0 threshold for the logits)
    mask_accurate = (mask_logits > 0.0) == gt_mask_logits.to(dtype=torch.uint8)
    mask_accurate = mask_accurate[~point_ignores]
    mask_accuracy = mask_accurate.nonzero().size(0) / max(mask_accurate.numel(), 1.0)
    get_event_storage().put_scalar("point/accuracy", mask_accuracy)

    point_loss = F.binary_cross_entropy_with_logits(
        mask_logits, gt_mask_logits.to(dtype=torch.float32), weight=~point_ignores, reduction="mean"
    )
    return point_loss


@POINT_HEAD_REGISTRY.register()
class StandardPointHead(nn.Module):
    """
    A point head multi-layer perceptron which we model with conv1d layers with kernel 1. The head
    takes both fine-grained and coarse prediction features as its input.
    """

    def __init__(self, cfg, input_shape: ShapeSpec):
        """
        The following attributes are parsed from config:
            fc_dim: the output dimension of each FC layers
            num_fc: the number of FC layers
            coarse_pred_each_layer: if True, coarse prediction features are concatenated to each
                layer's input
        """
        super(StandardPointHead, self).__init__()
        # fmt: off
        num_classes                 = cfg.MODEL.POINT_HEAD.NUM_CLASSES
        fc_dim                      = cfg.MODEL.POINT_HEAD.FC_DIM
        num_fc                      = cfg.MODEL.POINT_HEAD.NUM_FC
        cls_agnostic_mask           = cfg.MODEL.POINT_HEAD.CLS_AGNOSTIC_MASK
        self.coarse_pred_each_layer = cfg.MODEL.POINT_HEAD.COARSE_PRED_EACH_LAYER
        input_channels              = input_shape.channels
        # fmt: on

        fc_dim_in = input_channels + num_classes
        self.fc_layers = []
        for k in range(num_fc):
            fc = nn.Conv1d(fc_dim_in, fc_dim, kernel_size=1, stride=1, padding=0, bias=True)
            self.add_module("fc{}".format(k + 1), fc)
            self.fc_layers.append(fc)
            fc_dim_in = fc_dim
            fc_dim_in += num_classes if self.coarse_pred_each_layer else 0

        num_mask_classes = 1 if cls_agnostic_mask else num_classes
        self.predictor = nn.Conv1d(fc_dim_in, num_mask_classes, kernel_size=1, stride=1, padding=0)

        for layer in self.fc_layers:
            weight_init.c2_msra_fill(layer)
        # use normal distribution initialization for mask prediction layer
        nn.init.normal_(self.predictor.weight, std=0.001)
        if self.predictor.bias is not None:
            nn.init.constant_(self.predictor.bias, 0)

    def forward(self, fine_grained_features, coarse_features):
        x = torch.cat((fine_grained_features, coarse_features), dim=1)
        for layer in self.fc_layers:
            x = F.relu(layer(x))
            if self.coarse_pred_each_layer:
                x = cat((x, coarse_features), dim=1)
        return self.predictor(x)


@POINT_HEAD_REGISTRY.register()
class ImplicitPointHead(nn.Module):
    """
    A point head multi-layer perceptron which we model with conv1d layers with kernel 1. The head
    takes both fine-grained features and instance-wise MLP parameters as its input.
    """

    def __init__(self, cfg, input_shape: ShapeSpec):
        """
        The following attributes are parsed from config:
            channels: the output dimension of each FC layers
            num_layers: the number of FC layers (including the final prediction layer)
            image_feature_enabled: if True, fine-grained image-level features are used
            positional_encoding_enabled: if True, positional encoding is used
        """
        super(ImplicitPointHead, self).__init__()
        # fmt: off
        self.num_layers                         = cfg.MODEL.POINT_HEAD.NUM_FC + 1
        self.channels                           = cfg.MODEL.POINT_HEAD.FC_DIM
        self.image_feature_enabled              = cfg.MODEL.IMPLICIT_POINTREND.IMAGE_FEATURE_ENABLED
        self.positional_encoding_enabled        = cfg.MODEL.IMPLICIT_POINTREND.POS_ENC_ENABLED
        self.num_classes                        = (
            cfg.MODEL.POINT_HEAD.NUM_CLASSES if not cfg.MODEL.POINT_HEAD.CLS_AGNOSTIC_MASK else 1
        )
        self.in_channels                        = input_shape.channels
        # fmt: on

        if not self.image_feature_enabled:
            self.in_channels = 0
        if self.positional_encoding_enabled:
            self.in_channels += 256
            self.register_buffer("positional_encoding_gaussian_matrix", torch.randn((2, 128)))

        assert self.in_channels > 0

        num_weight_params, num_bias_params = [], []
        assert self.num_layers >= 2
        for l in range(self.num_layers):
            if l == 0:
                # input layer
                num_weight_params.append(self.in_channels * self.channels)
                num_bias_params.append(self.channels)
            elif l == self.num_layers - 1:
                # output layer
                num_weight_params.append(self.channels * self.num_classes)
                num_bias_params.append(self.num_classes)
            else:
                # intermediate layer
                num_weight_params.append(self.channels * self.channels)
                num_bias_params.append(self.channels)

        self.num_weight_params = num_weight_params
        self.num_bias_params = num_bias_params
        self.num_params = sum(num_weight_params) + sum(num_bias_params)

    def forward(self, fine_grained_features, point_coords, parameters):
        # features: [R, channels, K]
        # point_coords: [R, K, 2]
        num_instances = fine_grained_features.size(0)
        num_points = fine_grained_features.size(2)

        if num_instances == 0:
            return torch.zeros((0, 1, num_points), device=fine_grained_features.device)

        if self.positional_encoding_enabled:
            # locations: [R*K, 2]
            locations = 2 * point_coords.reshape(num_instances * num_points, 2) - 1
            locations = locations @ self.positional_encoding_gaussian_matrix.to(locations.device)
            locations = 2 * np.pi * locations
            locations = torch.cat([torch.sin(locations), torch.cos(locations)], dim=1)
            # locations: [R, C, K]
            locations = locations.reshape(num_instances, num_points, 256).permute(0, 2, 1)
            if not self.image_feature_enabled:
                fine_grained_features = locations
            else:
                fine_grained_features = torch.cat([locations, fine_grained_features], dim=1)

        # features [R, C, K]
        mask_feat = fine_grained_features.reshape(num_instances, self.in_channels, num_points)

        weights, biases = self._parse_params(
            parameters,
            self.in_channels,
            self.channels,
            self.num_classes,
            self.num_weight_params,
            self.num_bias_params,
        )

        point_logits = self._dynamic_mlp(mask_feat, weights, biases, num_instances)
        point_logits = point_logits.reshape(-1, self.num_classes, num_points)

        return point_logits

    @staticmethod
    def _dynamic_mlp(features, weights, biases, num_instances):
        assert features.dim() == 3, features.dim()
        n_layers = len(weights)
        x = features
        for i, (w, b) in enumerate(zip(weights, biases)):
            x = torch.einsum("nck,ndc->ndk", x, w) + b
            if i < n_layers - 1:
                x = F.relu(x)
        return x

    @staticmethod
    def _parse_params(
        pred_params,
        in_channels,
        channels,
        num_classes,
        num_weight_params,
        num_bias_params,
    ):
        assert pred_params.dim() == 2
        assert len(num_weight_params) == len(num_bias_params)
        assert pred_params.size(1) == sum(num_weight_params) + sum(num_bias_params)

        num_instances = pred_params.size(0)
        num_layers = len(num_weight_params)

        params_splits = list(
            torch.split_with_sizes(pred_params, num_weight_params + num_bias_params, dim=1)
        )

        weight_splits = params_splits[:num_layers]
        bias_splits = params_splits[num_layers:]

        for l in range(num_layers):
            if l == 0:
                # input layer
                weight_splits[l] = weight_splits[l].reshape(num_instances, channels, in_channels)
                bias_splits[l] = bias_splits[l].reshape(num_instances, channels, 1)
            elif l < num_layers - 1:
                # intermediate layer
                weight_splits[l] = weight_splits[l].reshape(num_instances, channels, channels)
                bias_splits[l] = bias_splits[l].reshape(num_instances, channels, 1)
            else:
                # output layer
                weight_splits[l] = weight_splits[l].reshape(num_instances, num_classes, channels)
                bias_splits[l] = bias_splits[l].reshape(num_instances, num_classes, 1)

        return weight_splits, bias_splits


def build_point_head(cfg, input_channels):
    """
    Build a point head defined by `cfg.MODEL.POINT_HEAD.NAME`.
    """
    head_name = cfg.MODEL.POINT_HEAD.NAME
    return POINT_HEAD_REGISTRY.get(head_name)(cfg, input_channels)