人工智能-项目实践-信息检索-使用深度学习网络（目标检测/特征提取/特征匹配）建立的图像精准检索系统

# Copyright 2016 Paul Balanca. 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. # ============================================================================== """Definition of 300 VGG-based SSD network. This model was initially introduced in: SSD: Single Shot MultiBox Detector Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg https://arxiv.org/abs/1512.02325 Two variants of the model are defined: the 300x300 and 512x512 models, the latter obtaining a slightly better accuracy on Pascal VOC. Usage: with slim.arg_scope(ssd_vgg.ssd_vgg()): outputs, end_points = ssd_vgg.ssd_vgg(inputs) This network port of the original Caffe model. The padding in TF and Caffe is slightly different, and can lead to severe accuracy drop if not taken care in a correct way! In Caffe, the output size of convolution and pooling layers are computing as following: h_o = (h_i + 2 * pad_h - kernel_h) / stride_h + 1 Nevertheless, there is a subtle difference between both for stride > 1. In the case of convolution: top_size = floor((bottom_size + 2*pad - kernel_size) / stride) + 1 whereas for pooling: top_size = ceil((bottom_size + 2*pad - kernel_size) / stride) + 1 Hence implicitely allowing some additional padding even if pad = 0. This behaviour explains why pooling with stride and kernel of size 2 are behaving the same way in TensorFlow and Caffe. Nevertheless, this is not the case anymore for other kernel sizes, hence motivating the use of special padding layer for controlling these side-effects. @@ssd_vgg_300 """ import math from collections import namedtuple import numpy as np import tensorflow as tf import tf_extended as tfe from nets import custom_layers from nets import ssd_common slim = tf.contrib.slim # =========================================================================== # # SSD class definition. # =========================================================================== # SSDParams = namedtuple('SSDParameters', ['img_shape', 'num_classes', 'no_annotation_label', 'feat_layers', 'feat_shapes', 'anchor_size_bounds', 'anchor_sizes', 'anchor_ratios', 'anchor_steps', 'anchor_offset', 'normalizations', 'prior_scaling' ]) class SSDNet(object): """Implementation of the SSD VGG-based 300 network. The default features layers with 300x300 image input are: conv4 ==> 38 x 38 conv7 ==> 19 x 19 conv8 ==> 10 x 10 conv9 ==> 5 x 5 conv10 ==> 3 x 3 conv11 ==> 1 x 1 The default image size used to train this network is 300x300. """ default_params = SSDParams( img_shape=(300, 300), num_classes=21, # 自己的类别+1 no_annotation_label=21, # 自己的类别+1 feat_layers=['block4', 'block7', 'block8', 'block9', 'block10', 'block11'], feat_shapes=[(38, 38), (19, 19), (10, 10), (5, 5), (3, 3), (1, 1)], anchor_size_bounds=[0.15, 0.90], # anchor_size_bounds=[0.20, 0.90], anchor_sizes=[(21., 45.), (45., 99.), (99., 153.), (153., 207.), (207., 261.), (261., 315.)], # anchor_sizes=[(30., 60.), # (60., 111.), # (111., 162.), # (162., 213.), # (213., 264.), # (264., 315.)], anchor_ratios=[[2, .5], [2, .5, 3, 1./3], [2, .5, 3, 1./3], [2, .5, 3, 1./3], [2, .5], [2, .5]], anchor_steps=[8, 16, 32, 64, 100, 300], anchor_offset=0.5, normalizations=[20, -1, -1, -1, -1, -1], prior_scaling=[0.1, 0.1, 0.2, 0.2] ) def __init__(self, params=None): """Init the SSD net with some parameters. Use the default ones if none provided. """ if isinstance(params, SSDParams): self.params = params else: self.params = SSDNet.default_params # ======================================================================= # def net(self, inputs, is_training=True, update_feat_shapes=True, dropout_keep_prob=0.5, prediction_fn=slim.softmax, reuse=None, scope='ssd_300_vgg'): """SSD network definition. """ r = ssd_net(inputs, num_classes=self.params.num_classes, feat_layers=self.params.feat_layers, anchor_sizes=self.params.anchor_sizes, anchor_ratios=self.params.anchor_ratios, normalizations=self.params.normalizations, is_training=is_training, dropout_keep_prob=dropout_keep_prob, prediction_fn=prediction_fn, reuse=reuse, scope=scope) # Update feature shapes (try at least!) if update_feat_shapes: shapes = ssd_feat_shapes_from_net(r[0], self.params.feat_shapes) self.params = self.params._replace(feat_shapes=shapes) return r def arg_scope(self, weight_decay=0.0005, data_format='NHWC'): """Network arg_scope. """ return ssd_arg_scope(weight_decay, data_format=data_format) def arg_scope_caffe(self, caffe_scope): """Caffe arg_scope used for weights importing. """ return ssd_arg_scope_caffe(caffe_scope) # ======================================================================= # def update_feature_shapes(self, predictions): """Update feature shapes from predictions collection (Tensor or Numpy array). """ shapes = ssd_feat_shapes_from_net(predictions, self.params.feat_shapes) self.params = self.params._replace(feat_shapes=shapes) def anchors(self, img_shape, dtype=np.float32): """Compute the default anchor boxes, given an image shape. """ return ssd_anchors_all_layers(img_shape, self.params.feat_shapes, self.params.anchor_sizes, self.params.anchor_ratios, self.params.anchor_steps, self.params.anchor_offset, dtype) def bboxes_encode(self, labels, bboxes, anchors, scope=None): """Encode labels and bounding boxes. """ return ssd_common.tf_ssd_bboxes_encode( labels, bboxes, anchors, self.params.num_classes, self