| # Licensed to the Apache Software Foundation (ASF) under one |
| # or more contributor license agreements. See the NOTICE file |
| # distributed with this work for additional information |
| # regarding copyright ownership. The ASF licenses this file |
| # to you 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 mxnet as mx |
| from . import proposal_target |
| |
| eps=2e-5 |
| use_global_stats=True |
| workspace=1024 |
| |
| |
| def residual_unit(data, num_filter, stride, dim_match, name): |
| bn1 = mx.sym.BatchNorm(data=data, fix_gamma=False, eps=eps, use_global_stats=use_global_stats, name=name + '_bn1') |
| act1 = mx.sym.Activation(data=bn1, act_type='relu', name=name + '_relu1') |
| conv1 = mx.sym.Convolution(data=act1, num_filter=int(num_filter * 0.25), kernel=(1, 1), stride=(1, 1), pad=(0, 0), |
| no_bias=True, workspace=workspace, name=name + '_conv1') |
| bn2 = mx.sym.BatchNorm(data=conv1, fix_gamma=False, eps=eps, use_global_stats=use_global_stats, name=name + '_bn2') |
| act2 = mx.sym.Activation(data=bn2, act_type='relu', name=name + '_relu2') |
| conv2 = mx.sym.Convolution(data=act2, num_filter=int(num_filter * 0.25), kernel=(3, 3), stride=stride, pad=(1, 1), |
| no_bias=True, workspace=workspace, name=name + '_conv2') |
| bn3 = mx.sym.BatchNorm(data=conv2, fix_gamma=False, eps=eps, use_global_stats=use_global_stats, name=name + '_bn3') |
| act3 = mx.sym.Activation(data=bn3, act_type='relu', name=name + '_relu3') |
| conv3 = mx.sym.Convolution(data=act3, num_filter=num_filter, kernel=(1, 1), stride=(1, 1), pad=(0, 0), no_bias=True, |
| workspace=workspace, name=name + '_conv3') |
| if dim_match: |
| shortcut = data |
| else: |
| shortcut = mx.sym.Convolution(data=act1, num_filter=num_filter, kernel=(1, 1), stride=stride, no_bias=True, |
| workspace=workspace, name=name + '_sc') |
| sum = mx.sym.ElementWiseSum(*[conv3, shortcut], name=name + '_plus') |
| return sum |
| |
| |
| def get_resnet_feature(data, units, filter_list): |
| # res1 |
| data_bn = mx.sym.BatchNorm(data=data, fix_gamma=True, eps=eps, use_global_stats=use_global_stats, name='bn_data') |
| conv0 = mx.sym.Convolution(data=data_bn, num_filter=64, kernel=(7, 7), stride=(2, 2), pad=(3, 3), |
| no_bias=True, name="conv0", workspace=workspace) |
| bn0 = mx.sym.BatchNorm(data=conv0, fix_gamma=False, eps=eps, use_global_stats=use_global_stats, name='bn0') |
| relu0 = mx.sym.Activation(data=bn0, act_type='relu', name='relu0') |
| pool0 = mx.symbol.Pooling(data=relu0, kernel=(3, 3), stride=(2, 2), pad=(1, 1), pool_type='max', name='pool0') |
| |
| # res2 |
| unit = residual_unit(data=pool0, num_filter=filter_list[0], stride=(1, 1), dim_match=False, name='stage1_unit1') |
| for i in range(2, units[0] + 1): |
| unit = residual_unit(data=unit, num_filter=filter_list[0], stride=(1, 1), dim_match=True, name='stage1_unit%s' % i) |
| |
| # res3 |
| unit = residual_unit(data=unit, num_filter=filter_list[1], stride=(2, 2), dim_match=False, name='stage2_unit1') |
| for i in range(2, units[1] + 1): |
| unit = residual_unit(data=unit, num_filter=filter_list[1], stride=(1, 1), dim_match=True, name='stage2_unit%s' % i) |
| |
| # res4 |
| unit = residual_unit(data=unit, num_filter=filter_list[2], stride=(2, 2), dim_match=False, name='stage3_unit1') |
| for i in range(2, units[2] + 1): |
| unit = residual_unit(data=unit, num_filter=filter_list[2], stride=(1, 1), dim_match=True, name='stage3_unit%s' % i) |
| return unit |
| |
| |
| def get_resnet_top_feature(data, units, filter_list): |
| unit = residual_unit(data=data, num_filter=filter_list[3], stride=(2, 2), dim_match=False, name='stage4_unit1') |
| for i in range(2, units[3] + 1): |
| unit = residual_unit(data=unit, num_filter=filter_list[3], stride=(1, 1), dim_match=True, name='stage4_unit%s' % i) |
| bn1 = mx.sym.BatchNorm(data=unit, fix_gamma=False, eps=eps, use_global_stats=use_global_stats, name='bn1') |
| relu1 = mx.sym.Activation(data=bn1, act_type='relu', name='relu1') |
| pool1 = mx.symbol.Pooling(data=relu1, global_pool=True, kernel=(7, 7), pool_type='avg', name='pool1') |
| return pool1 |
| |
| |
| def get_resnet_train(anchor_scales, anchor_ratios, rpn_feature_stride, |
| rpn_pre_topk, rpn_post_topk, rpn_nms_thresh, rpn_min_size, rpn_batch_rois, |
| num_classes, rcnn_feature_stride, rcnn_pooled_size, rcnn_batch_size, |
| rcnn_batch_rois, rcnn_fg_fraction, rcnn_fg_overlap, rcnn_bbox_stds, |
| units, filter_list): |
| num_anchors = len(anchor_scales) * len(anchor_ratios) |
| |
| data = mx.symbol.Variable(name="data") |
| im_info = mx.symbol.Variable(name="im_info") |
| gt_boxes = mx.symbol.Variable(name="gt_boxes") |
| rpn_label = mx.symbol.Variable(name='label') |
| rpn_bbox_target = mx.symbol.Variable(name='bbox_target') |
| rpn_bbox_weight = mx.symbol.Variable(name='bbox_weight') |
| |
| # shared convolutional layers |
| conv_feat = get_resnet_feature(data, units=units, filter_list=filter_list) |
| |
| # rpn feature |
| rpn_conv = mx.symbol.Convolution( |
| data=conv_feat, kernel=(3, 3), pad=(1, 1), num_filter=512, name="rpn_conv_3x3") |
| rpn_relu = mx.symbol.Activation(data=rpn_conv, act_type="relu", name="rpn_relu") |
| |
| # rpn classification |
| rpn_cls_score = mx.symbol.Convolution( |
| data=rpn_relu, kernel=(1, 1), pad=(0, 0), num_filter=2 * num_anchors, name="rpn_cls_score") |
| rpn_cls_score_reshape = mx.symbol.Reshape( |
| data=rpn_cls_score, shape=(0, 2, -1, 0), name="rpn_cls_score_reshape") |
| rpn_cls_prob = mx.symbol.SoftmaxOutput(data=rpn_cls_score_reshape, label=rpn_label, multi_output=True, |
| normalization='valid', use_ignore=True, ignore_label=-1, name="rpn_cls_prob") |
| rpn_cls_act = mx.symbol.softmax( |
| data=rpn_cls_score_reshape, axis=1, name="rpn_cls_act") |
| rpn_cls_act_reshape = mx.symbol.Reshape( |
| data=rpn_cls_act, shape=(0, 2 * num_anchors, -1, 0), name='rpn_cls_act_reshape') |
| |
| # rpn bbox regression |
| rpn_bbox_pred = mx.symbol.Convolution( |
| data=rpn_relu, kernel=(1, 1), pad=(0, 0), num_filter=4 * num_anchors, name="rpn_bbox_pred") |
| rpn_bbox_loss_ = rpn_bbox_weight * mx.symbol.smooth_l1(name='rpn_bbox_loss_', scalar=3.0, data=(rpn_bbox_pred - rpn_bbox_target)) |
| rpn_bbox_loss = mx.sym.MakeLoss(name='rpn_bbox_loss', data=rpn_bbox_loss_, grad_scale=1.0 / rpn_batch_rois) |
| |
| # rpn proposal |
| rois = mx.symbol.contrib.MultiProposal( |
| cls_prob=rpn_cls_act_reshape, bbox_pred=rpn_bbox_pred, im_info=im_info, name='rois', |
| feature_stride=rpn_feature_stride, scales=anchor_scales, ratios=anchor_ratios, |
| rpn_pre_nms_top_n=rpn_pre_topk, rpn_post_nms_top_n=rpn_post_topk, |
| threshold=rpn_nms_thresh, rpn_min_size=rpn_min_size) |
| |
| # rcnn roi proposal target |
| group = mx.symbol.Custom(rois=rois, gt_boxes=gt_boxes, op_type='proposal_target', |
| num_classes=num_classes, batch_images=rcnn_batch_size, |
| batch_rois=rcnn_batch_rois, fg_fraction=rcnn_fg_fraction, |
| fg_overlap=rcnn_fg_overlap, box_stds=rcnn_bbox_stds) |
| rois = group[0] |
| label = group[1] |
| bbox_target = group[2] |
| bbox_weight = group[3] |
| |
| # rcnn roi pool |
| roi_pool = mx.symbol.ROIPooling( |
| name='roi_pool', data=conv_feat, rois=rois, pooled_size=rcnn_pooled_size, spatial_scale=1.0 / rcnn_feature_stride) |
| |
| # rcnn top feature |
| top_feat = get_resnet_top_feature(roi_pool, units=units, filter_list=filter_list) |
| |
| # rcnn classification |
| cls_score = mx.symbol.FullyConnected(name='cls_score', data=top_feat, num_hidden=num_classes) |
| cls_prob = mx.symbol.SoftmaxOutput(name='cls_prob', data=cls_score, label=label, normalization='batch') |
| |
| # rcnn bbox regression |
| bbox_pred = mx.symbol.FullyConnected(name='bbox_pred', data=top_feat, num_hidden=num_classes * 4) |
| bbox_loss_ = bbox_weight * mx.symbol.smooth_l1(name='bbox_loss_', scalar=1.0, data=(bbox_pred - bbox_target)) |
| bbox_loss = mx.sym.MakeLoss(name='bbox_loss', data=bbox_loss_, grad_scale=1.0 / rcnn_batch_rois) |
| |
| # reshape output |
| label = mx.symbol.Reshape(data=label, shape=(rcnn_batch_size, -1), name='label_reshape') |
| cls_prob = mx.symbol.Reshape(data=cls_prob, shape=(rcnn_batch_size, -1, num_classes), name='cls_prob_reshape') |
| bbox_loss = mx.symbol.Reshape(data=bbox_loss, shape=(rcnn_batch_size, -1, 4 * num_classes), name='bbox_loss_reshape') |
| |
| # group output |
| group = mx.symbol.Group([rpn_cls_prob, rpn_bbox_loss, cls_prob, bbox_loss, mx.symbol.BlockGrad(label)]) |
| return group |
| |
| |
| def get_resnet_test(anchor_scales, anchor_ratios, rpn_feature_stride, |
| rpn_pre_topk, rpn_post_topk, rpn_nms_thresh, rpn_min_size, |
| num_classes, rcnn_feature_stride, rcnn_pooled_size, rcnn_batch_size, |
| units, filter_list): |
| num_anchors = len(anchor_scales) * len(anchor_ratios) |
| |
| data = mx.symbol.Variable(name="data") |
| im_info = mx.symbol.Variable(name="im_info") |
| |
| # shared convolutional layers |
| conv_feat = get_resnet_feature(data, units=units, filter_list=filter_list) |
| |
| # rpn feature |
| rpn_conv = mx.symbol.Convolution( |
| data=conv_feat, kernel=(3, 3), pad=(1, 1), num_filter=512, name="rpn_conv_3x3") |
| rpn_relu = mx.symbol.Activation(data=rpn_conv, act_type="relu", name="rpn_relu") |
| |
| # rpn classification |
| rpn_cls_score = mx.symbol.Convolution( |
| data=rpn_relu, kernel=(1, 1), pad=(0, 0), num_filter=2 * num_anchors, name="rpn_cls_score") |
| rpn_cls_score_reshape = mx.symbol.Reshape( |
| data=rpn_cls_score, shape=(0, 2, -1, 0), name="rpn_cls_score_reshape") |
| rpn_cls_act = mx.symbol.softmax( |
| data=rpn_cls_score_reshape, axis=1, name="rpn_cls_act") |
| rpn_cls_act_reshape = mx.symbol.Reshape( |
| data=rpn_cls_act, shape=(0, 2 * num_anchors, -1, 0), name='rpn_cls_act_reshape') |
| |
| # rpn bbox regression |
| rpn_bbox_pred = mx.symbol.Convolution( |
| data=rpn_relu, kernel=(1, 1), pad=(0, 0), num_filter=4 * num_anchors, name="rpn_bbox_pred") |
| |
| # rpn proposal |
| rois = mx.symbol.contrib.MultiProposal( |
| cls_prob=rpn_cls_act_reshape, bbox_pred=rpn_bbox_pred, im_info=im_info, name='rois', |
| feature_stride=rpn_feature_stride, scales=anchor_scales, ratios=anchor_ratios, |
| rpn_pre_nms_top_n=rpn_pre_topk, rpn_post_nms_top_n=rpn_post_topk, |
| threshold=rpn_nms_thresh, rpn_min_size=rpn_min_size) |
| |
| # rcnn roi pool |
| roi_pool = mx.symbol.ROIPooling( |
| name='roi_pool', data=conv_feat, rois=rois, pooled_size=rcnn_pooled_size, spatial_scale=1.0 / rcnn_feature_stride) |
| |
| # rcnn top feature |
| top_feat = get_resnet_top_feature(roi_pool, units=units, filter_list=filter_list) |
| |
| # rcnn classification |
| cls_score = mx.symbol.FullyConnected(name='cls_score', data=top_feat, num_hidden=num_classes) |
| cls_prob = mx.symbol.softmax(name='cls_prob', data=cls_score) |
| |
| # rcnn bbox regression |
| bbox_pred = mx.symbol.FullyConnected(name='bbox_pred', data=top_feat, num_hidden=num_classes * 4) |
| |
| # reshape output |
| cls_prob = mx.symbol.Reshape(data=cls_prob, shape=(rcnn_batch_size, -1, num_classes), name='cls_prob_reshape') |
| bbox_pred = mx.symbol.Reshape(data=bbox_pred, shape=(rcnn_batch_size, -1, 4 * num_classes), name='bbox_pred_reshape') |
| |
| # group output |
| group = mx.symbol.Group([rois, cls_prob, bbox_pred]) |
| return group |
