cascade_rcnn_cls_aware.py 12.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
# Copyright (c) 2019 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.

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import numpy as np
import sys

22 23 24
from collections import OrderedDict
import copy

25 26
import paddle.fluid as fluid
from ppdet.core.workspace import register
27
from .input_helper import multiscale_def
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

__all__ = ['CascadeRCNNClsAware']


@register
class CascadeRCNNClsAware(object):
    """
    Cascade R-CNN architecture, see https://arxiv.org/abs/1712.00726
    This is a kind of modification of Cascade R-CNN.
    Specifically, it predicts bboxes for all classes with different weights,
    while the standard vesion just predicts bboxes for foreground
    Args:
        backbone (object): backbone instance
        rpn_head (object): `RPNhead` instance
        bbox_assigner (object): `BBoxAssigner` instance
        roi_extractor (object): ROI extractor instance
        bbox_head (object): `BBoxHead` instance
        fpn (object): feature pyramid network instance
    """

    __category__ = 'architecture'
    __inject__ = [
        'backbone', 'fpn', 'rpn_head', 'bbox_assigner', 'roi_extractor',
        'bbox_head'
    ]

54 55 56 57 58 59 60 61
    def __init__(
            self,
            backbone,
            rpn_head,
            roi_extractor='FPNRoIAlign',
            bbox_head='CascadeBBoxHead',
            bbox_assigner='CascadeBBoxAssigner',
            fpn='FPN', ):
62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83
        super(CascadeRCNNClsAware, self).__init__()
        assert fpn is not None, "cascade RCNN requires FPN"
        self.backbone = backbone
        self.fpn = fpn
        self.rpn_head = rpn_head
        self.bbox_assigner = bbox_assigner
        self.roi_extractor = roi_extractor
        self.bbox_head = bbox_head
        self.bbox_clip = np.log(1000. / 16.)
        # Cascade local cfg
        (brw0, brw1, brw2) = self.bbox_assigner.bbox_reg_weights
        self.cascade_bbox_reg_weights = [
            [1. / brw0, 1. / brw0, 2. / brw0, 2. / brw0],
            [1. / brw1, 1. / brw1, 2. / brw1, 2. / brw1],
            [1. / brw2, 1. / brw2, 2. / brw2, 2. / brw2]
        ]
        self.cascade_rcnn_loss_weight = [1.0, 0.5, 0.25]

    def build(self, feed_vars, mode='train'):
        im = feed_vars['image']
        im_info = feed_vars['im_info']
        if mode == 'train':
84
            gt_bbox = feed_vars['gt_bbox']
85
            is_crowd = feed_vars['is_crowd']
86
            gt_class = feed_vars['gt_class']
87 88 89 90 91 92 93 94 95 96 97 98 99 100
        else:
            im_shape = feed_vars['im_shape']

        # backbone
        body_feats = self.backbone(im)

        # FPN
        if self.fpn is not None:
            body_feats, spatial_scale = self.fpn.get_output(body_feats)

        # rpn proposals
        rpn_rois = self.rpn_head.get_proposals(body_feats, im_info, mode=mode)

        if mode == 'train':
101
            rpn_loss = self.rpn_head.get_loss(im_info, gt_bbox, is_crowd)
102 103 104 105 106 107 108

        proposal_list = []
        roi_feat_list = []
        rcnn_pred_list = []
        rcnn_target_list = []

        bbox_pred = None
109

110 111
        self.cascade_var_v = []
        for stage in range(3):
112 113
            var_v = np.array(
                self.cascade_bbox_reg_weights[stage], dtype="float32")
114 115 116 117 118 119
            prior_box_var = fluid.layers.create_tensor(dtype="float32")
            fluid.layers.assign(input=var_v, output=prior_box_var)
            self.cascade_var_v.append(prior_box_var)

        self.cascade_decoded_box = []
        self.cascade_cls_prob = []
W
wangguanzhong 已提交
120
        max_overlap = None
121 122 123 124 125 126 127 128 129

        for stage in range(3):
            if stage > 0:
                pool_rois = decoded_assign_box
            else:
                pool_rois = rpn_rois
            if mode == "train":
                self.cascade_var_v[stage].stop_gradient = True
                outs = self.bbox_assigner(
W
wangguanzhong 已提交
130 131 132 133
                    input_rois=pool_rois,
                    feed_vars=feed_vars,
                    curr_stage=stage,
                    max_overlap=max_overlap)
134
                pool_rois = outs[0]
W
wangguanzhong 已提交
135 136
                max_overlap = outs[-1]
                rcnn_target_list.append(outs[:-1])
137

138 139 140
            # extract roi features
            roi_feat = self.roi_extractor(body_feats, pool_rois, spatial_scale)
            roi_feat_list.append(roi_feat)
141

142 143 144 145 146
            # bbox head
            cls_score, bbox_pred = self.bbox_head.get_output(
                roi_feat,
                cls_agnostic_bbox_reg=self.bbox_head.num_classes,
                wb_scalar=1.0 / self.cascade_rcnn_loss_weight[stage],
147
                name='_' + str(stage + 1))
148 149

            cls_prob = fluid.layers.softmax(cls_score, use_cudnn=False)
150

151
            decoded_box, decoded_assign_box = fluid.layers.box_decoder_and_assign(
152 153 154
                pool_rois, self.cascade_var_v[stage], bbox_pred, cls_prob,
                self.bbox_clip)

155 156 157 158
            if mode == "train":
                decoded_box.stop_gradient = True
                decoded_assign_box.stop_gradient = True
            else:
159
                self.cascade_cls_prob.append(cls_prob)
160
                self.cascade_decoded_box.append(decoded_box)
161

162
            rcnn_pred_list.append((cls_score, bbox_pred))
163

164 165
        # out loop
        if mode == 'train':
166
            loss = self.bbox_head.get_loss(rcnn_pred_list, rcnn_target_list,
167 168 169 170 171 172 173
                                           self.cascade_rcnn_loss_weight)
            loss.update(rpn_loss)
            total_loss = fluid.layers.sum(list(loss.values()))
            loss.update({'loss': total_loss})
            return loss
        else:
            pred = self.bbox_head.get_prediction_cls_aware(
174 175
                im_info, im_shape, self.cascade_cls_prob,
                self.cascade_decoded_box, self.cascade_bbox_reg_weights)
176
            return pred
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
    def build_multi_scale(self, feed_vars):
        required_fields = ['image', 'im_shape', 'im_info']
        self._input_check(required_fields, feed_vars)

        result = {}
        im_shape = feed_vars['im_shape']
        result['im_shape'] = im_shape

        for i in range(len(self.im_info_names) // 2):
            im = feed_vars[self.im_info_names[2 * i]]
            im_info = feed_vars[self.im_info_names[2 * i + 1]]

            # backbone
            body_feats = self.backbone(im)
            result.update(body_feats)
            # FPN
            if self.fpn is not None:
                body_feats, spatial_scale = self.fpn.get_output(body_feats)

            # rpn proposals
            rpn_rois = self.rpn_head.get_proposals(
                body_feats, im_info, mode="test")

            proposal_list = []
            roi_feat_list = []
            rcnn_pred_list = []
            rcnn_target_list = []

            bbox_pred = None

            self.cascade_var_v = []
            for stage in range(3):
                var_v = np.array(
                    self.cascade_bbox_reg_weights[stage], dtype="float32")
                prior_box_var = fluid.layers.create_tensor(dtype="float32")
                fluid.layers.assign(input=var_v, output=prior_box_var)
                self.cascade_var_v.append(prior_box_var)

            self.cascade_decoded_box = []
            self.cascade_cls_prob = []

            for stage in range(3):
                if stage > 0:
                    pool_rois = decoded_assign_box
                else:
                    pool_rois = rpn_rois

                # extract roi features
                roi_feat = self.roi_extractor(body_feats, pool_rois,
                                              spatial_scale)
                roi_feat_list.append(roi_feat)

                # bbox head
                cls_score, bbox_pred = self.bbox_head.get_output(
                    roi_feat,
                    cls_agnostic_bbox_reg=self.bbox_head.num_classes,
                    wb_scalar=1.0 / self.cascade_rcnn_loss_weight[stage],
                    name='_' + str(stage + 1))

                cls_prob = fluid.layers.softmax(cls_score, use_cudnn=False)

                decoded_box, decoded_assign_box = fluid.layers.box_decoder_and_assign(
                    pool_rois, self.cascade_var_v[stage], bbox_pred, cls_prob,
                    self.bbox_clip)

                self.cascade_cls_prob.append(cls_prob)
                self.cascade_decoded_box.append(decoded_box)

                rcnn_pred_list.append((cls_score, bbox_pred))

            pred = self.bbox_head.get_prediction_cls_aware(
                im_info,
                im_shape,
                self.cascade_cls_prob,
                self.cascade_decoded_box,
                self.cascade_bbox_reg_weights,
                return_box_score=True)

            bbox_name = 'bbox_' + str(i)
            score_name = 'score_' + str(i)
            if 'flip' in im.name:
                bbox_name += '_flip'
                score_name += '_flip'
            result[bbox_name] = pred['bbox']
            result[score_name] = pred['score']

        return result

266 267 268 269 270 271
    def _inputs_def(self, image_shape):
        im_shape = [None] + image_shape
        # yapf: disable
        inputs_def = {
            'image':    {'shape': im_shape,  'dtype': 'float32', 'lod_level': 0},
            'im_info':  {'shape': [None, 3], 'dtype': 'float32', 'lod_level': 0},
Q
qingqing01 已提交
272
            'im_id':    {'shape': [None, 1], 'dtype': 'int64',   'lod_level': 0},
273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
            'im_shape': {'shape': [None, 3], 'dtype': 'float32', 'lod_level': 0},
            'gt_bbox':  {'shape': [None, 4], 'dtype': 'float32', 'lod_level': 1},
            'gt_class': {'shape': [None, 1], 'dtype': 'int32',   'lod_level': 1},
            'is_crowd': {'shape': [None, 1], 'dtype': 'int32',   'lod_level': 1},
            'is_difficult': {'shape': [None, 1], 'dtype': 'int32', 'lod_level': 1},
        }
        # yapf: enable
        return inputs_def

    def build_inputs(self,
                     image_shape=[3, None, None],
                     fields=[
                         'image', 'im_info', 'im_id', 'gt_bbox', 'gt_class',
                         'is_crowd', 'gt_mask'
                     ],
288 289 290
                     multi_scale=False,
                     num_scales=-1,
                     use_flip=None,
291 292 293
                     use_dataloader=True,
                     iterable=False):
        inputs_def = self._inputs_def(image_shape)
294 295 296 297 298 299 300 301
        fields = copy.deepcopy(fields)
        if multi_scale:
            ms_def, ms_fields = multiscale_def(image_shape, num_scales,
                                               use_flip)
            inputs_def.update(ms_def)
            fields += ms_fields
            self.im_info_names = ['image', 'im_info'] + ms_fields

Y
Yang Zhang 已提交
302
        feed_vars = OrderedDict([(key, fluid.data(
303 304 305 306 307 308
            name=key,
            shape=inputs_def[key]['shape'],
            dtype=inputs_def[key]['dtype'],
            lod_level=inputs_def[key]['lod_level'])) for key in fields])
        loader = fluid.io.DataLoader.from_generator(
            feed_list=list(feed_vars.values()),
309
            capacity=16,
310 311 312
            use_double_buffer=True,
            iterable=iterable) if use_dataloader else None
        return feed_vars, loader
313

314 315 316 317 318
    def _input_check(self, require_fields, feed_vars):
        for var in require_fields:
            assert var in feed_vars, \
                "{} has no {} field".format(feed_vars, var)

319 320 321
    def train(self, feed_vars):
        return self.build(feed_vars, 'train')

322 323 324
    def eval(self, feed_vars, multi_scale=None):
        if multi_scale:
            return self.build_multi_scale(feed_vars)
325 326
        return self.build(feed_vars, 'test')

K
Kaipeng Deng 已提交
327
    def test(self, feed_vars, exclude_nms=False):
K
Kaipeng Deng 已提交
328 329
        assert not exclude_nms, "exclude_nms for {} is not support currently".format(
            self.__class__.__name__)
330
        return self.build(feed_vars, 'test')