wider face data 在 faster rcnn 上的实践记录(caffe)

按照githup上playerkk的工程进行实践，地址为：https://github.com/playerkk/face-py-faster-rcnn/blob/master/README.md

1.搭建faster rcnn

这个没什么好说的，参考rbg大神的官网即可，附上地址：https://github.com/rbgirshick/py-faster-rcnn

需要注意的是对于显卡是1080的，因为faster rcnn是基于老版本的cudnn，所以编译的时候会出现许多关于cudnn的报错问题。解决方案推荐的是用caffe中相关的文件替换掉faster rcnn中的相关文件。可参考地址：http://blog.csdn.net/u010733679/article/details/52221404。

2.克隆工程及下载预训练参数文件

git clone --recursive https://github.com/playerkk/face-py-faster-rcnn.git

在工程的根目录下执行：

cd face-py-faster-rcnn

./data/scripts/fetch_faster_rcnn_models.sh

会在data下出现 faster rcnn models.tgz。

3.下载wider face data数据

网站下载地址为：http://mmlab.ie.cuhk.edu.hk/projects/WIDERFace/

下载三个数据文件到一个目录中，我选择的是ubuntu的home目录下：

如图所示进行分布。

图中的最后一个文本文件从该网址下载：https://people.cs.umass.edu/~hzjiang//files/wider_face_train_annot.txt 。按照的是FDDB的格式。

在如图所示的data目录下建立symlinks：

会在工程的data目录下出现链接，如上图所示。

4.下载预训练的Imagenet模型

在工程目录下执行上图所示命令。

接下来就是开始执行训练的过程：

在工程的根目录下执行命令：

++++++++++++++++++++++++2017.0224++++++++++++++++++++更新+++++++++++++++++++++++++++

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

更新工程后我的train_net.py文件内容为：

#!/usr/bin/env python# --------------------------------------------------------# Fast R-CNN# Copyright (c) 2015 Microsoft# Licensed under The MIT License [see LICENSE for details]# Written by Ross Girshick# --------------------------------------------------------"""Train a Fast R-CNN network on a region of interest database."""# import matplotlib # matplotlib.use('Agg') import _init_pathsfrom fast_rcnn.train import get_training_roidb, train_netfrom fast_rcnn.config import cfg, cfg_from_file, cfg_from_list, get_output_dirfrom datasets.factory import get_imdbimport datasets.imdbimport caffeimport argparseimport pprintimport numpy as npimport sysdef parse_args():    """    Parse input arguments    """    parser = argparse.ArgumentParser(description='Train a Fast R-CNN network')    parser.add_argument('--gpu', dest='gpu_id',                        help='GPU device id to use [0]',                        default=0, type=int)    parser.add_argument('--solver', dest='solver',                        help='solver prototxt',                        default=None, type=str)    parser.add_argument('--iters', dest='max_iters',                        help='number of iterations to train',                        default=40000, type=int)    parser.add_argument('--weights', dest='pretrained_model',                        help='initialize with pretrained model weights',                        default=None, type=str)    parser.add_argument('--cfg', dest='cfg_file',                        help='optional config file',                        default=None, type=str)    parser.add_argument('--imdb', dest='imdb_name',                        help='dataset to train on',                        default='voc_2007_trainval', type=str)    parser.add_argument('--rand', dest='randomize',                        help='randomize (do not use a fixed seed)',                        action='store_true')    parser.add_argument('--set', dest='set_cfgs',                        help='set config keys', default=None,                        nargs=argparse.REMAINDER)    if len(sys.argv) == 1:        parser.print_help()        sys.exit(1)    args = parser.parse_args()    return argsdef combined_roidb(imdb_names):    def get_roidb(imdb_name):        imdb = get_imdb(imdb_name)        print 'Loaded dataset `{:s}` for training'.format(imdb.name)        imdb.set_proposal_method(cfg.TRAIN.PROPOSAL_METHOD)        print 'Set proposal method: {:s}'.format(cfg.TRAIN.PROPOSAL_METHOD)        roidb = get_training_roidb(imdb)        return roidb    roidbs = [get_roidb(s) for s in imdb_names.split('+')]    roidb = roidbs[0]    if len(roidbs) > 1:        for r in roidbs[1:]:            roidb.extend(r)        imdb = datasets.imdb.imdb(imdb_names)    else:        imdb = get_imdb(imdb_names)    return imdb, roidbif __name__ == '__main__':    args = parse_args()    print('Called with args:')    print(args)    if args.cfg_file is not None:        cfg_from_file(args.cfg_file)    if args.set_cfgs is not None:        cfg_from_list(args.set_cfgs)    cfg.GPU_ID = args.gpu_id    print('Using config:')    pprint.pprint(cfg)    if not args.randomize:        # fix the random seeds (numpy and caffe) for reproducibility        np.random.seed(cfg.RNG_SEED)        caffe.set_random_seed(cfg.RNG_SEED)    # set up caffe    caffe.set_mode_gpu()    caffe.set_device(args.gpu_id)    imdb, roidb = combined_roidb(args.imdb_name)    print '{:d} roidb entries'.format(len(roidb))    output_dir = get_output_dir(imdb)    print 'Output will be saved to `{:s}`'.format(output_dir)    train_net(args.solver, roidb, output_dir,              pretrained_model=args.pretrained_model,              max_iters=args.max_iters)

face.py:

# --------------------------------------------------------# Fast R-CNN# Copyright (c) 2015 Microsoft# Licensed under The MIT License [see LICENSE for details]# Written by Ross Girshick# --------------------------------------------------------# import datasets.face# import os# import datasets.imdb as imdb# import xml.dom.minidom as minidom# import numpy as np# import scipy.sparse# import scipy.io as sio# import utils.cython_bbox# import cPickle# import subprocessimport osfrom datasets.imdb import imdbimport datasets.ds_utils as ds_utilsimport xml.etree.ElementTree as ETimport numpy as npimport scipy.sparseimport scipy.io as sioimport utils.cython_bboximport cPickleimport subprocessimport uuidfrom voc_eval import voc_evalfrom fast_rcnn.config import cfgimport cv2import PILclass face(imdb):    def __init__(self, image_set, split, devkit_path):        imdb.__init__(self, 'wider')        self._image_set = image_set         # {'train', 'test'}        self._split = split                 # {1, 2, ..., 10}        self._devkit_path = devkit_path     # /data2/hzjiang/Data/CS2        # self._data_path = os.path.join(self._devkit_path, 'data')        self._data_path = self._devkit_path;        self._classes = ('__background__', # always index 0                         'face')        self._class_to_ind = dict(zip(self.classes, xrange(self.num_classes)))        self._image_ext = ['.png']        self._image_index, self._gt_roidb = self._load_image_set_index()        # Default to roidb handler        self._roidb_handler = self.selective_search_roidb        # Specific config options        self.config = {'cleanup'  : True,                       'use_salt' : True,                       'top_k'    : 2000}        assert os.path.exists(self._devkit_path), \                'Devkit path does not exist: {}'.format(self._devkit_path)        assert os.path.exists(self._data_path), \                'Path does not exist: {}'.format(self._data_path)    def image_path_at(self, i):        """        Return the absolute path to image i in the image sequence.        """        return self.image_path_from_index(self._image_index[i])    def image_path_from_index(self, index):        """        Construct an image path from the image's "index" identifier.        """        for ext in self._image_ext:            image_path = os.path.join(self._data_path, index)            if os.path.exists(image_path):                break        assert os.path.exists(image_path), \                'Path does not exist: {}'.format(image_path)        return image_path    def _load_image_set_index(self):        """        Load the indexes listed in this dataset's image set file.        """        # # Example path to image set file:        # # self._data_path + /ImageSets/val.txt        # # read from file        # image_set_file = 'split%d/%s_%d_annot.txt' % (self._fold, self._image_set, self._fold)        # # image_set_file = os.path.join(self._devkit_path, image_set_file)        # image_set_file = os.path.join('/home/hzjiang/Code/py-faster-rcnn/CS3-splits', image_set_file)        image_set_file = self._name + '_face_' + self._image_set + '_annot.txt'        image_set_file = os.path.join(self._devkit_path, image_set_file)        # image_set_file = 'cs3_rand_train_annot.txt'        # image_set_file = 'wider_dets_annot_from_cs3_model.txt'        # image_set_file = 'wider_manual_annot.txt'        assert os.path.exists(image_set_file), \                'Path does not exist: {}'.format(image_set_file)        image_index = []        gt_roidb = []                with open(image_set_file) as f:            # print len(f.lines())            lines = f.readlines()            idx = 0            while idx < len(lines):                image_name = lines[idx].split('\n')[0]                image_name = os.path.join('WIDER_%s/images' % self._image_set, image_name)                # print image_name                image_ext = os.path.splitext(image_name)[1].lower()                # print image_ext                assert(image_ext == '.png' or image_ext == '.jpg' or image_ext == '.jpeg')                image = PIL.Image.open(os.path.join(self._data_path, image_name))                imw = image.size[0]                imh = image.size[1]                idx += 1                num_boxes = int(lines[idx])                # print num_boxes                boxes = np.zeros((num_boxes, 4), dtype=np.uint16)                gt_classes = np.zeros((num_boxes), dtype=np.int32)                overlaps = np.zeros((num_boxes, self.num_classes), dtype=np.float32)                for i in xrange(num_boxes):                    idx += 1                    coor = map(float, lines[idx].split())                    x1 = min(max(coor[0], 0), imw - 1)                    y1 = min(max(coor[1], 0), imh - 1)                    x2 = min(max(x1 + coor[2] - 1, 0), imw - 1)                    y2 = min(max(y1 + coor[3] - 1, 0), imh - 1)                    if np.isnan(x1):                        x1 = -1                    if np.isnan(y1):                        y1 = -1                    if np.isnan(x2):                        x2 = -1                    if np.isnan(y2):                        y2 = -1                                            cls = self._class_to_ind['face']                    boxes[i, :] = [x1, y1, x2, y2]                    gt_classes[i] = cls                    overlaps[i, cls] = 1.0                widths = boxes[:, 2] - boxes[:, 0] + 1                heights = boxes[:, 3] - boxes[:, 1] + 1                keep_idx = np.where(np.bitwise_and(widths > 5, heights > 5))                if len(keep_idx[0]) <= 0:                    idx += 1                    continue                boxes = boxes[keep_idx]                gt_classes = gt_classes[keep_idx[0]]                overlaps = overlaps[keep_idx[0], :]                if not (boxes[:, 2] >= boxes[:, 0]).all():                    print boxes                    print image_name                # print boxes                assert (boxes[:, 2] >= boxes[:, 0]).all()                assert (boxes[:, 3] >= boxes[:, 1]).all()                overlaps = scipy.sparse.csr_matrix(overlaps)                gt_roidb.append({'boxes' : boxes,                                'gt_classes': gt_classes,                                'gt_overlaps' : overlaps,                                'flipped' : False,                                'image_name': image_name})                image_index.append(image_name)                idx += 1                    assert(idx == len(lines))        return image_index, gt_roidb    def gt_roidb(self):        """        Return the database of ground-truth regions of interest.        This function loads/saves from/to a cache file to speed up future calls.        """        cache_file = os.path.join(self.cache_path, self.name + '_gt_roidb.pkl')        if os.path.exists(cache_file):            with open(cache_file, 'rb') as fid:                roidb = cPickle.load(fid)            print '{} gt roidb loaded from {}'.format(self.name, cache_file)            return roidb        with open(cache_file, 'wb') as fid:            cPickle.dump(self._gt_roidb, fid, cPickle.HIGHEST_PROTOCOL)        print 'wrote gt roidb to {}'.format(cache_file)        return self._gt_roidb    def selective_search_roidb(self):        """        Return the database of selective search regions of interest.        Ground-truth ROIs are also included.        This function loads/saves from/to a cache file to speed up future calls.        """        cache_file = os.path.join(self.cache_path,                                  self.name + '_selective_search_roidb.pkl')        if os.path.exists(cache_file):            with open(cache_file, 'rb') as fid:                roidb = cPickle.load(fid)            print '{} ss roidb loaded from {}'.format(self.name, cache_file)            return roidb        if self._image_set != 'test':            gt_roidb = self.gt_roidb()            ss_roidb = self._load_selective_search_roidb(gt_roidb)            roidb = datasets.imdb.merge_roidbs(gt_roidb, ss_roidb)        else:            roidb = self._load_selective_search_roidb(None)            print len(roidb)        with open(cache_file, 'wb') as fid:            cPickle.dump(roidb, fid, cPickle.HIGHEST_PROTOCOL)        print 'wrote ss roidb to {}'.format(cache_file)        return roidb    def _load_selective_search_roidb(self, gt_roidb):        filename = os.path.abspath(os.path.join(self._devkit_path,                                                self.name + '.mat'))        assert os.path.exists(filename), \               'Selective search data not found at: {}'.format(filename)        raw_data = sio.loadmat(filename)['all_boxes'].ravel()        box_list = []        for i in xrange(raw_data.shape[0]):            boxes = raw_data[i][:, (1, 0, 3, 2)] - 1            assert (boxes[:, 2] >= boxes[:, 0]).all()            box_list.append(boxes)        return self.create_roidb_from_box_list(box_list, gt_roidb)    def selective_search_IJCV_roidb(self):        """        Return the database of selective search regions of interest.        Ground-truth ROIs are also included.        This function loads/saves from/to a cache file to speed up future calls.        """        cache_file = os.path.join(self.cache_path,                '{:s}_selective_search_IJCV_top_{:d}_roidb.pkl'.                format(self.name, self.config['top_k']))        if os.path.exists(cache_file):            with open(cache_file, 'rb') as fid:                roidb = cPickle.load(fid)            print '{} ss roidb loaded from {}'.format(self.name, cache_file)            return roidb        gt_roidb = self.gt_roidb()        ss_roidb = self._load_selective_search_IJCV_roidb(gt_roidb)        roidb = datasets.imdb.merge_roidbs(gt_roidb, ss_roidb)        with open(cache_file, 'wb') as fid:            cPickle.dump(roidb, fid, cPickle.HIGHEST_PROTOCOL)        print 'wrote ss roidb to {}'.format(cache_file)        return roidb    def _load_selective_search_IJCV_roidb(self, gt_roidb):        IJCV_path = os.path.abspath(os.path.join(self.cache_path, '..',                                                 'selective_search_IJCV_data',                                                 self.name))        assert os.path.exists(IJCV_path), \               'Selective search IJCV data not found at: {}'.format(IJCV_path)        top_k = self.config['top_k']        box_list = []        for i in xrange(self.num_images):            filename = os.path.join(IJCV_path, self.image_index[i] + '.mat')            raw_data = sio.loadmat(filename)            box_list.append((raw_data['boxes'][:top_k, :]-1).astype(np.uint16))        return self.create_roidb_from_box_list(box_list, gt_roidb)    def _load_face_annotation(self, index):        """        Load image and bounding boxes info from txt files of face.        """        filename = os.path.join(self._data_path, 'Annotations', index + '.mat')        data = sio.loadmat(filename)        num_objs = data['gt'].shape[0]        boxes = np.zeros((num_objs, 4), dtype=np.uint16)        gt_classes = np.zeros((num_objs), dtype=np.int32)        overlaps = np.zeros((num_objs, self.num_classes), dtype=np.float32)        # Load object bounding boxes into a data frame.        for ix in xrange(num_objs):            # Make pixel indexes 0-based            coor = data['gt'][ix, :]            x1 = float(coor[0]) - 1            y1 = float(coor[1]) - 1            x2 = float(coor[2]) - 1            y2 = float(coor[3]) - 1            cls = self._class_to_ind['face']            boxes[ix, :] = [x1, y1, x2, y2]            gt_classes[ix] = cls            overlaps[ix, cls] = 1.0        overlaps = scipy.sparse.csr_matrix(overlaps)        if not (boxes[:, 2] >= boxes[:, 0]).all():            print boxes            print filename        assert (boxes[:, 2] >= boxes[:, 0]).all()        return {'boxes' : boxes,                'gt_classes': gt_classes,                'gt_overlaps' : overlaps,                'flipped' : False}    def _write_inria_results_file(self, all_boxes):        use_salt = self.config['use_salt']        comp_id = 'comp4'        if use_salt:            comp_id += '-{}'.format(os.getpid())        # VOCdevkit/results/comp4-44503_det_test_aeroplane.txt        path = os.path.join(self._devkit_path, 'results', self.name, comp_id + '_')        for cls_ind, cls in enumerate(self.classes):            if cls == '__background__':                continue            print 'Writing {} results file'.format(cls)            filename = path + 'det_' + self._image_set + '_' + cls + '.txt'            with open(filename, 'wt') as f:                for im_ind, index in enumerate(self.image_index):                    dets = all_boxes[cls_ind][im_ind]                    if dets == []:                        continue                    # the VOCdevkit expects 1-based indices                    for k in xrange(dets.shape[0]):                        f.write('{:s} {:.3f} {:.1f} {:.1f} {:.1f} {:.1f}\n'.                                format(index, dets[k, -1],                                       dets[k, 0] + 1, dets[k, 1] + 1,                                       dets[k, 2] + 1, dets[k, 3] + 1))        return comp_id    def _do_matlab_eval(self, comp_id, output_dir='output'):        rm_results = self.config['cleanup']        path = os.path.join(os.path.dirname(__file__),                            'VOCdevkit-matlab-wrapper')        cmd = 'cd {} && '.format(path)        cmd += '{:s} -nodisplay -nodesktop '.format(datasets.MATLAB)        cmd += '-r "dbstop if error; '        cmd += 'setenv(\'LC_ALL\',\'C\'); voc_eval(\'{:s}\',\'{:s}\',\'{:s}\',\'{:s}\',{:d}); quit;"' \               .format(self._devkit_path, comp_id,                       self._image_set, output_dir, int(rm_results))        print('Running:\n{}'.format(cmd))        status = subprocess.call(cmd, shell=True)    def evaluate_detections(self, all_boxes, output_dir):        comp_id = self._write_inria_results_file(all_boxes)        self._do_matlab_eval(comp_id, output_dir)    def competition_mode(self, on):        if on:            self.config['use_salt'] = False            self.config['cleanup'] = False        else:            self.config['use_salt'] = True            self.config['cleanup'] = Trueif __name__ == '__main__':    d = datasets.inria('train', '')    res = d.roidb    from IPython import embed; embed()

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

之后在命令行中输入：

voole@zhx:~/face-py-faster-rcnn$ ./experiments/scripts/faster_rcnn_end2end.sh 0 VGG16 wider

输出：

问题已经得到解决了：原因是因为我安装的easydict的版本是1.4，即使我配置了faster_rcnn_end2end.yml，但是easydict并没有将参数传给train_net.py，所以才会出现上述问题，解决方案就是升级你的easydict。