生活随笔
收集整理的這篇文章主要介紹了
R-FCN+ResNet-50用自己的数据集训练模型(python版本)
小編覺得挺不錯(cuò)的,現(xiàn)在分享給大家,幫大家做個(gè)參考.
說明：
本文假設(shè)你已經(jīng)做好數(shù)據(jù)集，格式和VOC2007一致，并且Linux系統(tǒng)已經(jīng)配置好caffe所需環(huán)境（博客里教程很多），下面是訓(xùn)練的一些修改。
py-R-FCN源碼下載地址：
https://github.com/Orpine/py-R-FCN
也有Matlab版本：
https://github.com/daijifeng001/R-FCN
本文用到的是python版本。
本文主要參考https://github.com/Orpine/py-R-FCN。
準(zhǔn)備工作：
（1）配置caffe環(huán)境(網(wǎng)上找教程)
（2）安裝cython,?python-opencv,?easydict
[plain] view plaincopy
pip?install?cython??pip?install?easydict??apt-get?install?python-opencv??
然后，我們就可以開始配置R-FCN了。
1.下載py-R-FCN
[plain] view plaincopy
git?clone?https://github.com/Orpine/py-R-FCN.git??
下面稱你的py-R-FCN路徑為RFCN_ROOT.
2.下載caffe
注意，該caffe版本是微軟版本
[plain] view plaincopy
cd?$RFCN_ROOT??git?clone?https://github.com/Microsoft/caffe.git??如果一切正常的話，python代碼會(huì)自動(dòng)添加環(huán)境變量 $RFCN_ROOT/caffe/python，否則，你需要自己添加環(huán)境變量。
3.Build Cython
[plain] view plaincopy
cd?$RFCN_ROOT/lib??make??4.Build caffe和pycaffe
[plain] view plaincopy
cd?$RFCN_ROOT/caffe??cp?Makefile.config.example?Makefile.config??然后修改Makefile.config。caffe必須支持python層，所以WITH_PYTHON_LAYER := 1是必須的。其他配置可參考：Makefile.config
接著：
[plain] view plaincopy
cd?$RFCN_ROOT/caffe??make?-j8?&&?make?pycaffe??如果沒有出錯(cuò)，則：
5.測試Demo
經(jīng)過上面的工作，我們可以測試一下是否可以正常運(yùn)行。
我們需要下載作者訓(xùn)練好的模型，地址：鏈接：http://pan.baidu.com/s/1kVGy8DL 密碼：pwwg
然后將模型放在$RFCN_ROOT/data。看起來是這樣的：
$RFCN_ROOT/data/rfcn_models/resnet50_rfcn_final.caffemodel
$RFCN_ROOT/data/rfcn_models/resnet101_rfcn_final.caffemodel運(yùn)行：
[plain] view plaincopy
cd?$RFCN_ROOT??./tools/demo_rfcn.py?--net?ResNet-50??6.用我們的數(shù)據(jù)集訓(xùn)練
（1）拷貝數(shù)據(jù)集
假設(shè)我們已經(jīng)做好數(shù)據(jù)集了，格式是和VOC2007一致，將你的數(shù)據(jù)集
拷貝到$RFCN_ROOT/data下?？雌饋硎沁@樣的：
$VOCdevkit0712/
# development kit
$VOCdevkit/VOCcode/
# VOC utility code
$VOCdevkit/VOC0712
# image sets, annotations, etc.
# ... and several other directories ...如果你的文件夾名字不是VOCdevkit0712和VOC0712，修改成0712就行了。
（作者是用VOC2007和VOC2012訓(xùn)練的，所以文件夾名字帶0712。也可以修改代碼，但是那樣比較麻煩一些，修改文件夾比較簡單）
（2）下載預(yù)訓(xùn)練模型
本文以ResNet-50為例，因此下載ResNet-50-model.caffemodel。下載地址：鏈接：http://pan.baidu.com/s/1slRHD0L 密碼：r3ki
然后將caffemodel放在$RFCN_ROOT/data/imagenet_models ?(data下沒有該文件夾就新建一個(gè))
（3）修改模型網(wǎng)絡(luò)
打開$RFCN_ROOT/models/pascal_voc/ResNet-50/rfcn_end2end ?(以end2end為例)
注意：下面的cls_num指的是你數(shù)據(jù)集的類別數(shù)+1（背景）。比如我有15類，+1類背景，cls_num=16.
<1>修改class-aware/train_ohem.prototxt
[plain] view plaincopy
layer?{????name:?'input-data'????type:?'Python'????top:?'data'????top:?'im_info'????top:?'gt_boxes'????python_param?{??????module:?'roi_data_layer.layer'??????layer:?'RoIDataLayer'??????param_str:?"'num_classes':?16"?#cls_num????}??}??[plain] view plaincopy
layer?{????name:?'roi-data'????type:?'Python'????bottom:?'rpn_rois'????bottom:?'gt_boxes'????top:?'rois'????top:?'labels'????top:?'bbox_targets'????top:?'bbox_inside_weights'????top:?'bbox_outside_weights'????python_param?{??????module:?'rpn.proposal_target_layer'??????layer:?'ProposalTargetLayer'??????param_str:?"'num_classes':?16"?#cls_num????}??}??[plain] view plaincopy
layer?{??????bottom:?"conv_new_1"??????top:?"rfcn_cls"??????name:?"rfcn_cls"??????type:?"Convolution"??????convolution_param?{??????????num_output:?784?#cls_num*(score_maps_size^2)??????????kernel_size:?1??????????pad:?0??????????weight_filler?{??????????????type:?"gaussian"??????????????std:?0.01??????????}??????????bias_filler?{??????????????type:?"constant"??????????????value:?0??????????}??????}??????param?{??????????lr_mult:?1.0??????}??????param?{??????????lr_mult:?2.0??????}??}??[plain] view plaincopy
layer?{??????bottom:?"conv_new_1"??????top:?"rfcn_bbox"??????name:?"rfcn_bbox"??????type:?"Convolution"??????convolution_param?{??????????num_output:?3136?#4*cls_num*(score_maps_size^2)??????????kernel_size:?1??????????pad:?0??????????weight_filler?{??????????????type:?"gaussian"??????????????std:?0.01??????????}??????????bias_filler?{??????????????type:?"constant"??????????????value:?0??????????}??????}??????param?{??????????lr_mult:?1.0??????}??????param?{??????????lr_mult:?2.0??????}??}??[plain] view plaincopy
layer?{??????bottom:?"rfcn_cls"??????bottom:?"rois"??????top:?"psroipooled_cls_rois"??????name:?"psroipooled_cls_rois"??????type:?"PSROIPooling"??????psroi_pooling_param?{??????????spatial_scale:?0.0625??????????output_dim:?16??#cls_num??????????group_size:?7??????}??}??
[plain] view plaincopy
layer?{??????bottom:?"rfcn_bbox"??????bottom:?"rois"??????top:?"psroipooled_loc_rois"??????name:?"psroipooled_loc_rois"??????type:?"PSROIPooling"??????psroi_pooling_param?{??????????spatial_scale:?0.0625??????????output_dim:?64?#4*cls_num??????????group_size:?7??????}??}??
<2>修改class-aware/test.prototxt
[plain] view plaincopy
layer?{??????bottom:?"conv_new_1"??????top:?"rfcn_cls"??????name:?"rfcn_cls"??????type:?"Convolution"??????convolution_param?{??????????num_output:?784?#cls_num*(score_maps_size^2)??????????kernel_size:?1??????????pad:?0??????????weight_filler?{??????????????type:?"gaussian"??????????????std:?0.01??????????}??????????bias_filler?{??????????????type:?"constant"??????????????value:?0??????????}??????}??????param?{??????????lr_mult:?1.0??????}??????param?{??????????lr_mult:?2.0??????}??}??[plain] view plaincopy
layer?{??????bottom:?"conv_new_1"??????top:?"rfcn_bbox"??????name:?"rfcn_bbox"??????type:?"Convolution"??????convolution_param?{??????????num_output:?3136?#4*cls_num*(score_maps_size^2)??????????kernel_size:?1??????????pad:?0??????????weight_filler?{??????????????type:?"gaussian"??????????????std:?0.01??????????}??????????bias_filler?{??????????????type:?"constant"??????????????value:?0??????????}??????}??????param?{??????????lr_mult:?1.0??????}??????param?{??????????lr_mult:?2.0??????}??}??[plain] view plaincopy
layer?{??????bottom:?"rfcn_cls"??????bottom:?"rois"??????top:?"psroipooled_cls_rois"??????name:?"psroipooled_cls_rois"??????type:?"PSROIPooling"??????psroi_pooling_param?{??????????spatial_scale:?0.0625??????????output_dim:?16??#cls_num??????????group_size:?7??????}??}??[plain] view plaincopy
layer?{??????bottom:?"rfcn_bbox"??????bottom:?"rois"??????top:?"psroipooled_loc_rois"??????name:?"psroipooled_loc_rois"??????type:?"PSROIPooling"??????psroi_pooling_param?{??????????spatial_scale:?0.0625??????????output_dim:?64??#4*cls_num??????????group_size:?7??????}??}??[plain] view plaincopy
layer?{??????name:?"cls_prob_reshape"??????type:?"Reshape"??????bottom:?"cls_prob_pre"??????top:?"cls_prob"??????reshape_param?{??????????shape?{??????????????dim:?-1??????????????dim:?16??#cls_num??????????}??????}??}??[plain] view plaincopy
layer?{??????name:?"bbox_pred_reshape"??????type:?"Reshape"??????bottom:?"bbox_pred_pre"??????top:?"bbox_pred"??????reshape_param?{??????????shape?{??????????????dim:?-1??????????????dim:?64??#4*cls_num??????????}??????}??}??
<3>修改train_agnostic.prototxt
[plain] view plaincopy
layer?{????name:?'input-data'????type:?'Python'????top:?'data'????top:?'im_info'????top:?'gt_boxes'????python_param?{??????module:?'roi_data_layer.layer'??????layer:?'RoIDataLayer'??????param_str:?"'num_classes':?16"??#cls_num????}??}??[plain] view plaincopy
layer?{??????bottom:?"conv_new_1"??????top:?"rfcn_cls"??????name:?"rfcn_cls"??????type:?"Convolution"??????convolution_param?{??????????num_output:?784?#cls_num*(score_maps_size^2)???###??????????kernel_size:?1??????????pad:?0??????????weight_filler?{??????????????type:?"gaussian"??????????????std:?0.01??????????}??????????bias_filler?{??????????????type:?"constant"??????????????value:?0??????????}??????}??????param?{??????????lr_mult:?1.0??????}??????param?{??????????lr_mult:?2.0??????}??}??[plain] view plaincopy
layer?{??????bottom:?"rfcn_cls"??????bottom:?"rois"??????top:?"psroipooled_cls_rois"??????name:?"psroipooled_cls_rois"??????type:?"PSROIPooling"??????psroi_pooling_param?{??????????spatial_scale:?0.0625??????????output_dim:?16?#cls_num???###??????????group_size:?7??????}??}??<4>修改train_agnostic_ohem.prototxt
[plain] view plaincopy
layer?{????name:?'input-data'????type:?'Python'????top:?'data'????top:?'im_info'????top:?'gt_boxes'????python_param?{??????module:?'roi_data_layer.layer'??????layer:?'RoIDataLayer'??????param_str:?"'num_classes':?16"?#cls_num?###????}??}??[plain] view plaincopy
layer?{??????bottom:?"conv_new_1"??????top:?"rfcn_cls"??????name:?"rfcn_cls"??????type:?"Convolution"??????convolution_param?{??????????num_output:?784?#cls_num*(score_maps_size^2)???###??????????kernel_size:?1??????????pad:?0??????????weight_filler?{??????????????type:?"gaussian"??????????????std:?0.01??????????}??????????bias_filler?{??????????????type:?"constant"??????????????value:?0??????????}??????}??????param?{??????????lr_mult:?1.0??????}??????param?{??????????lr_mult:?2.0??????}??}??[plain] view plaincopy
layer?{??????bottom:?"rfcn_cls"??????bottom:?"rois"??????top:?"psroipooled_cls_rois"??????name:?"psroipooled_cls_rois"??????type:?"PSROIPooling"??????psroi_pooling_param?{??????????spatial_scale:?0.0625??????????output_dim:?16?#cls_num???###??????????group_size:?7??????}??}??
<5>修改test_agnostic.prototxt
[plain] view plaincopy
layer?{??????bottom:?"conv_new_1"??????top:?"rfcn_cls"??????name:?"rfcn_cls"??????type:?"Convolution"??????convolution_param?{??????????num_output:?784?#cls_num*(score_maps_size^2)?###??????????kernel_size:?1??????????pad:?0??????????weight_filler?{??????????????type:?"gaussian"??????????????std:?0.01??????????}??????????bias_filler?{??????????????type:?"constant"??????????????value:?0??????????}??????}??????param?{??????????lr_mult:?1.0??????}??????param?{??????????lr_mult:?2.0??????}??}??[plain] view plaincopy
layer?{??????bottom:?"rfcn_cls"??????bottom:?"rois"??????top:?"psroipooled_cls_rois"??????name:?"psroipooled_cls_rois"??????type:?"PSROIPooling"??????psroi_pooling_param?{??????????spatial_scale:?0.0625??????????output_dim:?16?#cls_num???###??????????group_size:?7??????}??}??[plain] view plaincopy
layer?{??????name:?"cls_prob_reshape"??????type:?"Reshape"??????bottom:?"cls_prob_pre"??????top:?"cls_prob"??????reshape_param?{??????????shape?{??????????????dim:?-1??????????????dim:?16?#cls_num???###??????????}??????}??}??(4)修改代碼
<1>$RFCN/lib/datasets/pascal_voc.py
[plain] view plaincopy
class?pascal_voc(imdb):??????def?__init__(self,?image_set,?year,?devkit_path=None):??????????imdb.__init__(self,?'voc_'?+?year?+?'_'?+?image_set)??????????self._year?=?year??????????self._image_set?=?image_set??????????self._devkit_path?=?self._get_default_path()?if?devkit_path?is?None?\??????????????????????????????else?devkit_path??????????self._data_path?=?os.path.join(self._devkit_path,?'VOC'?+?self._year)??????????self._classes?=?('__background__',?#?always?index?0???????????????????????????'你的標(biāo)簽1','你的標(biāo)簽2',你的標(biāo)簽3','你的標(biāo)簽4'????????????????????????)??改成你的數(shù)據(jù)集標(biāo)簽。
<2>$RFCN_ROOT/lib/datasets/imdb.py
主要是assert (boxes[:, 2] >= boxes[:, 0]).all()可能出現(xiàn)AssertionError，具體解決辦法參考：
http://blog.csdn.net/xzzppp/article/details/52036794
PS：
上面將有無ohem的prototxt都改了，但是這里訓(xùn)練用的是ohem。
另外，默認(rèn)的迭代次數(shù)很大，可以修改$RFCN\experiments\scripts\rfcn_end2end_ohem.sh:
[plain] view plaincopy
case?$DATASET?in????pascal_voc)??????TRAIN_IMDB="voc_0712_trainval"??????TEST_IMDB="voc_0712_test"??????PT_DIR="pascal_voc"??????ITERS=110000??修改ITERS為你想要的迭代次數(shù)即可。
（5）開始訓(xùn)練
[plain] view plaincopy
cd?$RFCN_ROOT??./experiments/scripts/rfcn_end2end_ohem.sh?0?ResNet-50?pascal_voc??正常的話，就開始迭代了：
$
RFCN_ROOT/experiments/scripts里還有一些其他的訓(xùn)練方法，也可以測試一下（經(jīng)過上面的修改，無ohem的end2end訓(xùn)練也改好了，其他訓(xùn)練方法修改的過程差不多）。
（6）結(jié)果
將訓(xùn)練得到的模型($RFCN_ROOT/output/rfcn_end2end_ohem/voc_0712_trainval里最后的caffemodel)拷貝到$RFCN_ROOT/data/rfcn_models下，然后打開$RFCN_ROOT/tools/demo_rfcn.py，將CLASSES修改成你的標(biāo)簽，NETS修改成你的model，im_names修改成你的測試圖片(放在data/demo下),最后：
[plain] view plaincopy
cd?$RFCN_ROOT??./tools/demo_rfcn.py?--net?ResNet-50??
我將顯示的標(biāo)簽改為了中文，修改方法參考：
http://blog.csdn.net/sinat_30071459/article/details/51694037
總結(jié)
以上是生活随笔為你收集整理的R-FCN+ResNet-50用自己的数据集训练模型(python版本)的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。
如果覺得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。
