ppyoloe_head.py 16.3 KB
Newer Older
S
shangliang Xu 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
# Copyright (c) 2022 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.

import paddle
import paddle.nn as nn
import paddle.nn.functional as F
from ppdet.core.workspace import register

from ..bbox_utils import batch_distance2bbox
from ..losses import GIoULoss
from ..initializer import bias_init_with_prob, constant_, normal_
from ..assigners.utils import generate_anchors_for_grid_cell
from ppdet.modeling.backbones.cspresnet import ConvBNLayer
W
wangguanzhong 已提交
25
from ppdet.modeling.ops import get_static_shape, get_act_fn
W
wangxinxin08 已提交
26
from ppdet.modeling.layers import MultiClassNMS
S
shangliang Xu 已提交
27

S
shangliang Xu 已提交
28
__all__ = ['PPYOLOEHead']
S
shangliang Xu 已提交
29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47


class ESEAttn(nn.Layer):
    def __init__(self, feat_channels, act='swish'):
        super(ESEAttn, self).__init__()
        self.fc = nn.Conv2D(feat_channels, feat_channels, 1)
        self.conv = ConvBNLayer(feat_channels, feat_channels, 1, act=act)

        self._init_weights()

    def _init_weights(self):
        normal_(self.fc.weight, std=0.001)

    def forward(self, feat, avg_feat):
        weight = F.sigmoid(self.fc(avg_feat))
        return self.conv(feat * weight)


@register
S
shangliang Xu 已提交
48
class PPYOLOEHead(nn.Layer):
49
    __shared__ = [
50 51
        'num_classes', 'eval_size', 'trt', 'exclude_nms',
        'exclude_post_process', 'use_shared_conv'
52
    ]
S
shangliang Xu 已提交
53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
    __inject__ = ['static_assigner', 'assigner', 'nms']

    def __init__(self,
                 in_channels=[1024, 512, 256],
                 num_classes=80,
                 act='swish',
                 fpn_strides=(32, 16, 8),
                 grid_cell_scale=5.0,
                 grid_cell_offset=0.5,
                 reg_max=16,
                 static_assigner_epoch=4,
                 use_varifocal_loss=True,
                 static_assigner='ATSSAssigner',
                 assigner='TaskAlignedAssigner',
                 nms='MultiClassNMS',
68
                 eval_size=None,
S
shangliang Xu 已提交
69 70 71 72 73
                 loss_weight={
                     'class': 1.0,
                     'iou': 2.5,
                     'dfl': 0.5,
                 },
S
shangliang Xu 已提交
74
                 trt=False,
75
                 exclude_nms=False,
76 77
                 exclude_post_process=False,
                 use_shared_conv=True):
S
shangliang Xu 已提交
78
        super(PPYOLOEHead, self).__init__()
S
shangliang Xu 已提交
79 80 81 82 83 84 85 86 87 88
        assert len(in_channels) > 0, "len(in_channels) should > 0"
        self.in_channels = in_channels
        self.num_classes = num_classes
        self.fpn_strides = fpn_strides
        self.grid_cell_scale = grid_cell_scale
        self.grid_cell_offset = grid_cell_offset
        self.reg_max = reg_max
        self.iou_loss = GIoULoss()
        self.loss_weight = loss_weight
        self.use_varifocal_loss = use_varifocal_loss
89
        self.eval_size = eval_size
S
shangliang Xu 已提交
90 91 92 93 94

        self.static_assigner_epoch = static_assigner_epoch
        self.static_assigner = static_assigner
        self.assigner = assigner
        self.nms = nms
W
wangxinxin08 已提交
95 96
        if isinstance(self.nms, MultiClassNMS) and trt:
            self.nms.trt = trt
S
shangliang Xu 已提交
97
        self.exclude_nms = exclude_nms
98
        self.exclude_post_process = exclude_post_process
99 100
        self.use_shared_conv = use_shared_conv

S
shangliang Xu 已提交
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
        # stem
        self.stem_cls = nn.LayerList()
        self.stem_reg = nn.LayerList()
        act = get_act_fn(
            act, trt=trt) if act is None or isinstance(act,
                                                       (str, dict)) else act
        for in_c in self.in_channels:
            self.stem_cls.append(ESEAttn(in_c, act=act))
            self.stem_reg.append(ESEAttn(in_c, act=act))
        # pred head
        self.pred_cls = nn.LayerList()
        self.pred_reg = nn.LayerList()
        for in_c in self.in_channels:
            self.pred_cls.append(
                nn.Conv2D(
                    in_c, self.num_classes, 3, padding=1))
            self.pred_reg.append(
                nn.Conv2D(
                    in_c, 4 * (self.reg_max + 1), 3, padding=1))
        # projection conv
        self.proj_conv = nn.Conv2D(self.reg_max + 1, 1, 1, bias_attr=False)
122
        self.proj_conv.skip_quant = True
S
shangliang Xu 已提交
123 124 125 126 127 128 129 130 131 132 133 134 135 136
        self._init_weights()

    @classmethod
    def from_config(cls, cfg, input_shape):
        return {'in_channels': [i.channels for i in input_shape], }

    def _init_weights(self):
        bias_cls = bias_init_with_prob(0.01)
        for cls_, reg_ in zip(self.pred_cls, self.pred_reg):
            constant_(cls_.weight)
            constant_(cls_.bias, bias_cls)
            constant_(reg_.weight)
            constant_(reg_.bias, 1.0)

137 138 139
        proj = paddle.linspace(0, self.reg_max, self.reg_max + 1).reshape(
            [1, self.reg_max + 1, 1, 1])
        self.proj_conv.weight.set_value(proj)
S
shangliang Xu 已提交
140
        self.proj_conv.weight.stop_gradient = True
141
        if self.eval_size:
S
shangliang Xu 已提交
142
            anchor_points, stride_tensor = self._generate_anchors()
W
wangxinxin08 已提交
143 144
            self.anchor_points = anchor_points
            self.stride_tensor = stride_tensor
S
shangliang Xu 已提交
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

    def forward_train(self, feats, targets):
        anchors, anchor_points, num_anchors_list, stride_tensor = \
            generate_anchors_for_grid_cell(
                feats, self.fpn_strides, self.grid_cell_scale,
                self.grid_cell_offset)

        cls_score_list, reg_distri_list = [], []
        for i, feat in enumerate(feats):
            avg_feat = F.adaptive_avg_pool2d(feat, (1, 1))
            cls_logit = self.pred_cls[i](self.stem_cls[i](feat, avg_feat) +
                                         feat)
            reg_distri = self.pred_reg[i](self.stem_reg[i](feat, avg_feat))
            # cls and reg
            cls_score = F.sigmoid(cls_logit)
            cls_score_list.append(cls_score.flatten(2).transpose([0, 2, 1]))
            reg_distri_list.append(reg_distri.flatten(2).transpose([0, 2, 1]))
        cls_score_list = paddle.concat(cls_score_list, axis=1)
        reg_distri_list = paddle.concat(reg_distri_list, axis=1)

        return self.get_loss([
            cls_score_list, reg_distri_list, anchors, anchor_points,
            num_anchors_list, stride_tensor
        ], targets)

S
shangliang Xu 已提交
170
    def _generate_anchors(self, feats=None, dtype='float32'):
S
shangliang Xu 已提交
171 172 173 174 175 176 177
        # just use in eval time
        anchor_points = []
        stride_tensor = []
        for i, stride in enumerate(self.fpn_strides):
            if feats is not None:
                _, _, h, w = feats[i].shape
            else:
178 179
                h = int(self.eval_size[0] / stride)
                w = int(self.eval_size[1] / stride)
S
shangliang Xu 已提交
180 181 182 183 184
            shift_x = paddle.arange(end=w) + self.grid_cell_offset
            shift_y = paddle.arange(end=h) + self.grid_cell_offset
            shift_y, shift_x = paddle.meshgrid(shift_y, shift_x)
            anchor_point = paddle.cast(
                paddle.stack(
S
shangliang Xu 已提交
185
                    [shift_x, shift_y], axis=-1), dtype=dtype)
S
shangliang Xu 已提交
186
            anchor_points.append(anchor_point.reshape([-1, 2]))
S
shangliang Xu 已提交
187
            stride_tensor.append(paddle.full([h * w, 1], stride, dtype=dtype))
S
shangliang Xu 已提交
188 189 190 191 192
        anchor_points = paddle.concat(anchor_points)
        stride_tensor = paddle.concat(stride_tensor)
        return anchor_points, stride_tensor

    def forward_eval(self, feats):
193
        if self.eval_size:
S
shangliang Xu 已提交
194 195 196 197 198
            anchor_points, stride_tensor = self.anchor_points, self.stride_tensor
        else:
            anchor_points, stride_tensor = self._generate_anchors(feats)
        cls_score_list, reg_dist_list = [], []
        for i, feat in enumerate(feats):
199
            _, _, h, w = feat.shape
S
shangliang Xu 已提交
200 201 202 203 204 205
            l = h * w
            avg_feat = F.adaptive_avg_pool2d(feat, (1, 1))
            cls_logit = self.pred_cls[i](self.stem_cls[i](feat, avg_feat) +
                                         feat)
            reg_dist = self.pred_reg[i](self.stem_reg[i](feat, avg_feat))
            reg_dist = reg_dist.reshape([-1, 4, self.reg_max + 1, l]).transpose(
206
                [0, 2, 3, 1])
207 208 209 210 211
            if self.use_shared_conv:
                reg_dist = self.proj_conv(F.softmax(
                    reg_dist, axis=1)).squeeze(1)
            else:
                reg_dist = F.softmax(reg_dist, axis=1)
S
shangliang Xu 已提交
212 213
            # cls and reg
            cls_score = F.sigmoid(cls_logit)
214
            cls_score_list.append(cls_score.reshape([-1, self.num_classes, l]))
215
            reg_dist_list.append(reg_dist)
S
shangliang Xu 已提交
216 217

        cls_score_list = paddle.concat(cls_score_list, axis=-1)
218 219 220 221 222
        if self.use_shared_conv:
            reg_dist_list = paddle.concat(reg_dist_list, axis=1)
        else:
            reg_dist_list = paddle.concat(reg_dist_list, axis=2)
            reg_dist_list = self.proj_conv(reg_dist_list).squeeze(1)
S
shangliang Xu 已提交
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

        return cls_score_list, reg_dist_list, anchor_points, stride_tensor

    def forward(self, feats, targets=None):
        assert len(feats) == len(self.fpn_strides), \
            "The size of feats is not equal to size of fpn_strides"

        if self.training:
            return self.forward_train(feats, targets)
        else:
            return self.forward_eval(feats)

    @staticmethod
    def _focal_loss(score, label, alpha=0.25, gamma=2.0):
        weight = (score - label).pow(gamma)
        if alpha > 0:
            alpha_t = alpha * label + (1 - alpha) * (1 - label)
            weight *= alpha_t
        loss = F.binary_cross_entropy(
            score, label, weight=weight, reduction='sum')
        return loss

    @staticmethod
    def _varifocal_loss(pred_score, gt_score, label, alpha=0.75, gamma=2.0):
        weight = alpha * pred_score.pow(gamma) * (1 - label) + gt_score * label
        loss = F.binary_cross_entropy(
            pred_score, gt_score, weight=weight, reduction='sum')
        return loss

    def _bbox_decode(self, anchor_points, pred_dist):
253 254
        _, l, _ = get_static_shape(pred_dist)
        pred_dist = F.softmax(pred_dist.reshape([-1, l, 4, self.reg_max + 1]))
255
        pred_dist = self.proj_conv(pred_dist.transpose([0, 3, 1, 2])).squeeze(1)
S
shangliang Xu 已提交
256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309
        return batch_distance2bbox(anchor_points, pred_dist)

    def _bbox2distance(self, points, bbox):
        x1y1, x2y2 = paddle.split(bbox, 2, -1)
        lt = points - x1y1
        rb = x2y2 - points
        return paddle.concat([lt, rb], -1).clip(0, self.reg_max - 0.01)

    def _df_loss(self, pred_dist, target):
        target_left = paddle.cast(target, 'int64')
        target_right = target_left + 1
        weight_left = target_right.astype('float32') - target
        weight_right = 1 - weight_left
        loss_left = F.cross_entropy(
            pred_dist, target_left, reduction='none') * weight_left
        loss_right = F.cross_entropy(
            pred_dist, target_right, reduction='none') * weight_right
        return (loss_left + loss_right).mean(-1, keepdim=True)

    def _bbox_loss(self, pred_dist, pred_bboxes, anchor_points, assigned_labels,
                   assigned_bboxes, assigned_scores, assigned_scores_sum):
        # select positive samples mask
        mask_positive = (assigned_labels != self.num_classes)
        num_pos = mask_positive.sum()
        # pos/neg loss
        if num_pos > 0:
            # l1 + iou
            bbox_mask = mask_positive.unsqueeze(-1).tile([1, 1, 4])
            pred_bboxes_pos = paddle.masked_select(pred_bboxes,
                                                   bbox_mask).reshape([-1, 4])
            assigned_bboxes_pos = paddle.masked_select(
                assigned_bboxes, bbox_mask).reshape([-1, 4])
            bbox_weight = paddle.masked_select(
                assigned_scores.sum(-1), mask_positive).unsqueeze(-1)

            loss_l1 = F.l1_loss(pred_bboxes_pos, assigned_bboxes_pos)

            loss_iou = self.iou_loss(pred_bboxes_pos,
                                     assigned_bboxes_pos) * bbox_weight
            loss_iou = loss_iou.sum() / assigned_scores_sum

            dist_mask = mask_positive.unsqueeze(-1).tile(
                [1, 1, (self.reg_max + 1) * 4])
            pred_dist_pos = paddle.masked_select(
                pred_dist, dist_mask).reshape([-1, 4, self.reg_max + 1])
            assigned_ltrb = self._bbox2distance(anchor_points, assigned_bboxes)
            assigned_ltrb_pos = paddle.masked_select(
                assigned_ltrb, bbox_mask).reshape([-1, 4])
            loss_dfl = self._df_loss(pred_dist_pos,
                                     assigned_ltrb_pos) * bbox_weight
            loss_dfl = loss_dfl.sum() / assigned_scores_sum
        else:
            loss_l1 = paddle.zeros([1])
            loss_iou = paddle.zeros([1])
310
            loss_dfl = pred_dist.sum() * 0.
S
shangliang Xu 已提交
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
        return loss_l1, loss_iou, loss_dfl

    def get_loss(self, head_outs, gt_meta):
        pred_scores, pred_distri, anchors,\
        anchor_points, num_anchors_list, stride_tensor = head_outs

        anchor_points_s = anchor_points / stride_tensor
        pred_bboxes = self._bbox_decode(anchor_points_s, pred_distri)

        gt_labels = gt_meta['gt_class']
        gt_bboxes = gt_meta['gt_bbox']
        pad_gt_mask = gt_meta['pad_gt_mask']
        # label assignment
        if gt_meta['epoch_id'] < self.static_assigner_epoch:
            assigned_labels, assigned_bboxes, assigned_scores = \
                self.static_assigner(
                    anchors,
                    num_anchors_list,
                    gt_labels,
                    gt_bboxes,
                    pad_gt_mask,
                    bg_index=self.num_classes,
                    pred_bboxes=pred_bboxes.detach() * stride_tensor)
            alpha_l = 0.25
        else:
            assigned_labels, assigned_bboxes, assigned_scores = \
                self.assigner(
                pred_scores.detach(),
                pred_bboxes.detach() * stride_tensor,
                anchor_points,
                num_anchors_list,
                gt_labels,
                gt_bboxes,
                pad_gt_mask,
                bg_index=self.num_classes)
            alpha_l = -1
        # rescale bbox
        assigned_bboxes /= stride_tensor
        # cls loss
        if self.use_varifocal_loss:
S
shangliang Xu 已提交
351 352
            one_hot_label = F.one_hot(assigned_labels,
                                      self.num_classes + 1)[..., :-1]
S
shangliang Xu 已提交
353 354 355
            loss_cls = self._varifocal_loss(pred_scores, assigned_scores,
                                            one_hot_label)
        else:
S
shangliang Xu 已提交
356
            loss_cls = self._focal_loss(pred_scores, assigned_scores, alpha_l)
S
shangliang Xu 已提交
357 358

        assigned_scores_sum = assigned_scores.sum()
W
wangguanzhong 已提交
359
        if paddle.distributed.get_world_size() > 1:
S
shangliang Xu 已提交
360
            paddle.distributed.all_reduce(assigned_scores_sum)
361 362
            assigned_scores_sum /= paddle.distributed.get_world_size()
        assigned_scores_sum = paddle.clip(assigned_scores_sum, min=1.)
S
shangliang Xu 已提交
363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
        loss_cls /= assigned_scores_sum

        loss_l1, loss_iou, loss_dfl = \
            self._bbox_loss(pred_distri, pred_bboxes, anchor_points_s,
                            assigned_labels, assigned_bboxes, assigned_scores,
                            assigned_scores_sum)
        loss = self.loss_weight['class'] * loss_cls + \
               self.loss_weight['iou'] * loss_iou + \
               self.loss_weight['dfl'] * loss_dfl
        out_dict = {
            'loss': loss,
            'loss_cls': loss_cls,
            'loss_iou': loss_iou,
            'loss_dfl': loss_dfl,
            'loss_l1': loss_l1,
        }
        return out_dict

S
shangliang Xu 已提交
381
    def post_process(self, head_outs, scale_factor):
S
shangliang Xu 已提交
382
        pred_scores, pred_dist, anchor_points, stride_tensor = head_outs
383
        pred_bboxes = batch_distance2bbox(anchor_points, pred_dist)
S
shangliang Xu 已提交
384
        pred_bboxes *= stride_tensor
385 386 387
        if self.exclude_post_process:
            return paddle.concat(
                [pred_bboxes, pred_scores.transpose([0, 2, 1])], axis=-1), None
S
shangliang Xu 已提交
388
        else:
389 390 391 392 393 394 395 396 397 398 399 400
            # scale bbox to origin
            scale_y, scale_x = paddle.split(scale_factor, 2, axis=-1)
            scale_factor = paddle.concat(
                [scale_x, scale_y, scale_x, scale_y],
                axis=-1).reshape([-1, 1, 4])
            pred_bboxes /= scale_factor
            if self.exclude_nms:
                # `exclude_nms=True` just use in benchmark
                return pred_bboxes, pred_scores
            else:
                bbox_pred, bbox_num, _ = self.nms(pred_bboxes, pred_scores)
                return bbox_pred, bbox_num