使用kaist数据集训练tf-faster-rcnn

使用kaist数据集训练tf-faster-rcnn
首先是修改pascal_voc.py,新建自己的kaist_rgb.py
pacal_voc的数据格式很麻烦，annotation是xml格式，自己的数据一般都是txt,所以，不会像读取pascal_voc数据那么复杂。下面是前人在pascal_voc.py 基础上写的kaist Dataset的接口：

# --------------------------------------------------------
# Fast R-CNN
# Copyright (c) 2015 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Ross Girshick 
# --------------------------------------------------------#import datasets.caltech
import os
from datasets.imdb import imdb
import xml.dom.minidom as minidom
import numpy as np
import scipy.sparse
import scipy.io as sio
import utils.cython_bbox
import pickle
import subprocess
from model.config import cfgclass kaist_rgb(imdb):def __init__(self, image_set):imdb.__init__(self, 'kaist_' + image_set)  # image_set: train04 or testself._image_set = image_setself._devkit_path = self._get_default_path()self._data_path = self._get_default_path()self._classes = ('__background__', 'pedestrian')self._class_to_ind = dict(zip(self.classes, range(self.num_classes)))# self._class_to_ind = {'__background__': 0, 'pedestrian': 1}self._image_ext = '.jpg'self._image_index = self._load_image_set_index()# Default to roidb handlerself._roidb_handler = self.selective_search_roidb# PASCAL specific config optionsself.config = {'cleanup': True,'use_salt': True,'use_diff': False,'matlab_eval': False,'rpn_file': None,'min_size': 2}assert os.path.exists(self._devkit_path), 'VOCdevkit 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."""# image_path = os.path.join(self._data_path, self._image_set, 'images', index + self._image_ext)image_path = os.path.join(self._data_path, self._image_set, 'images', index[:-6] + 'visible/' + index[-6:] + self._image_ext)assert os.path.exists(image_path), 'Path does not exist: {}'.format(image_path)return image_pathdef _load_image_set_index(self):"""Load the indexes listed in this dataset's image set file."""# Example path to image set file:# self._devkit_path + /VOCdevkit2007/VOC2007/ImageSets/Main/val.txtimage_set_file = os.path.join(self._data_path, self._image_set, self._image_set + '.txt')assert os.path.exists(image_set_file), 'Path does not exist: {}'.format(image_set_file)with open(image_set_file) as f:image_index = [x.strip() for x in f.readlines()]return image_indexdef _get_default_path(self):"""Return the default path where kaist dataset is expected to be installed."""return os.path.join(cfg.DATA_DIR, 'kaist')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:try:roidb = pickle.load(fid)except:roidb = pickle.load(fid, encoding='bytes')print ('{} gt roidb loaded from {}'.format(self.name, cache_file))#print (roidb)#for dic in roidb:#    print (dic['gt_overlaps'])return roidbgt_roidb = [self._load_revised_annotation(index)for index in self.image_index]#print (gt_roidb)with open(cache_file, 'wb') as fid:pickle.dump(gt_roidb, fid, pickle.HIGHEST_PROTOCOL)print ('wrote gt roidb to {}'.format(cache_file))return gt_roidbdef 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 = pickle.load(fid)print ('{} ss roidb loaded from {}'.format(self.name, cache_file))return roidbif self._image_set != 'test-all':gt_roidb = self.gt_roidb()ss_roidb = self._load_selective_search_roidb(gt_roidb)roidb = imdb.merge_roidbs(gt_roidb, ss_roidb)else:roidb = self._load_selective_search_roidb(None)with open(cache_file, 'wb') as fid:pickle.dump(roidb, fid, pickle.HIGHEST_PROTOCOL)print('wrote ss roidb to {}'.format(cache_file))return roidbdef _load_selective_search_roidb(self, gt_roidb):filename = os.path.abspath(os.path.join(self.cache_path, '..','selective_search_data',self.name + '.mat'))assert os.path.exists(filename), 'Selective search data not found at: {}'.format(filename)raw_data = sio.loadmat(filename)['boxes'].ravel()box_list = []for i in xrange(raw_data.shape[0]):box_list.append(raw_data[i][:, :] - 1)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 = pickle.load(fid)print ('{} ss roidb loaded from {}'.format(self.name, cache_file))gt_roidb = self.gt_roidb()ss_roidb = self._load_selective_search_IJCV_roidb(gt_roidb)roidb = imdb.merge_roidbs(gt_roidb, ss_roidb)with open(cache_file, 'wb') as fid:pickle.dump(roidb, fid, pickle.HIGHEST_PROTOCOL)print ('wrote ss roidb to {}'.format(cache_file))return roidbdef rpn_roidb(self):if self._image_set != 'test-all':gt_roidb = self.gt_roidb()rpn_roidb = self._load_rpn_roidb(gt_roidb)roidb = imdb.merge_roidbs(gt_roidb, rpn_roidb)else:roidb = self._load_rpn_roidb(None)return roidbdef _load_rpn_roidb(self, gt_roidb):filename = self.config['rpn_file']print ('loading {}'.format(filename))assert os.path.exists(filename), 'rpn data not found at: {}'.format(filename)with open(filename, 'rb') as f:box_list = pickle.load(f)return self.create_roidb_from_box_list(box_list, gt_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','voc_' + self._year))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_revised_annotation(self, index):"""Load image and bounding boxes info from text file in the kaist dataset format."""filename = os.path.join(self._data_path, self._image_set, 'annotations', index + '.txt')# print ('Loading: {}'.format(filename))with open(filename) as f:lines = f.readlines()[1:]num_objs = len(lines)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)seg_areas = np.zeros((num_objs), dtype=np.float32)# Load object bounding boxes into a data frame.ix = 0for obj in lines:# Make pixel indexes 0-basedinfo = obj.split()# jamif self._image_set.find("train") != -1:if info[0] == "person":x1 = float(info[1])y1 = float(info[2])x2 = x1 + float(info[3])y2 = y1 + float(info[4])assert(x2 >= x1)assert(y2 >= y1)cls = self._class_to_ind['pedestrian']boxes[ix, :] = [max(x1 - 1, 0), max(y1 - 1, 0), min(x2 - 1, 639), min(y2 - 1, 479)]gt_classes[ix] = clsoverlaps[ix, cls] = 1.0seg_areas[ix] = (x2 - x1 + 1) * (y2 - y1 + 1)ix = ix + 1overlaps = scipy.sparse.csr_matrix(overlaps)return {'boxes': boxes,'gt_classes': gt_classes,'gt_overlaps': overlaps,'flipped': False,'seg_areas': seg_areas}def _write_voc_results_file(self, all_boxes):use_salt = self.config['use_salt']comp_id = 'comp4'if use_salt:comp_id += '-{}'.format(os.getpid())# VOCdevkit/results/VOC2007/Main/comp4-44503_det_test_aeroplane.txtpath = os.path.join(self._devkit_path, 'results', 'VOC', 'Main', comp_id + '_')for cls_ind, cls in enumerate(self.classes):if cls == '__background__':continueprint ('Writing {} VOC 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 indicesfor 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_iddef _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 += '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_voc_results_file(all_boxes)self._do_matlab_eval(comp_id, output_dir)def competition_mode(self, on):if on:self.config['use_salt'] = Falseself.config['cleanup'] = Falseelse:self.config['use_salt'] = Trueself.config['cleanup'] = Trueif __name__ == '__main__':d = datasets.kaist('train20')res = d.roidbfrom IPython import embedembed()

2018-09-06 13:11:49.699197: W tensorflow/core/framework/op_kernel.cc:1263] Invalid argument: ValueError: attempt to get argmax of an empty sequence
Traceback (most recent call last):File "/home/ramsey/.local/lib/python3.5/site-packages/tensorflow/python/ops/script_ops.py", line 206, in __call__ret = func(*args)File "/home/ramsey/tf-faster-rcnn/tools/../lib/layer_utils/anchor_target_layer.py", line 57, in anchor_target_layerargmax_overlaps = overlaps.argmax(axis=1)ValueError: attempt to get argmax of an empty sequence

这里附上数据文件的组织形式。

通过输出roidd的部分信息以及计算得到的overlaps，发现，validation roidb的gt_boxes是空的。·（如下图）

最开始没有找到原因，于是注释掉了lib/model/train_val.py中train_model方法的check validation data部分，想先跑通训练。（如下图）

注释掉以后，成功跑通训练。单在迭代一定次数后，又出现了之前的问题：

2018-09-06 13:11:49.699197: W tensorflow/core/framework/op_kernel.cc:1263] Invalid argument: ValueError: attempt to get argmax of an empty sequence
Traceback (most recent call last):File "/home/ramsey/.local/lib/python3.5/site-packages/tensorflow/python/ops/script_ops.py", line 206, in __call__ret = func(*args)File "/home/ramsey/tf-faster-rcnn/tools/../lib/layer_utils/anchor_target_layer.py", line 57, in anchor_target_layerargmax_overlaps = overlaps.argmax(axis=1)ValueError: attempt to get argmax of an empty sequence

于是，尝试输出用于training的roidb（是filt 空gt box之后的roidb），发现存在部分roidb的gt_box的四个坐标是0。

所以，从这里猜想，应该是数据出了问题。
根据输出日志，找到相应的数据例如，上面的I02759.jpg对应的annotation

这个标注很奇怪！因为其他正确的gt_box的annotation都是person.
于是，找到KAIST DATASET的原论文（Multispectral Pedestrian Detection: Benchmark Dataset and Baseline）
其中有提到：”Obviously an individual pedestrian was labelled as a person. Not distinguishable individuals were labeled as people. People riding a two-wheeled vehicle were labeled as cyclist. In a highly cluttered scene, even human annotators sometimes cannot clearly determine whether a human shaped object is a pedestrian or not. This object is labeled as person? an it is ignored in the validation“
所以，KAIST 数据集的annotation不光只有person,还有cyclist, person? people。

而在kaist_rgb.py中，load_annotation函数，只读取了annotation是person的数据，但在创建box的numy的时候，又根据的是读取的数据的行数：boxes = np.zeros((num_objs, 4), dtype=np.uint16), 所以导致部分gt_boxes的存在但四个坐标都是0. 这样是为什么明明有filter_roidb函数，但仍然出错的原因。因为tf_faster_rcnn的filter_roidb(roidb)函数只能去掉gt_box为空的roidb,不能去掉gt_box的四个坐标都为0的roidb

  def _load_revised_annotation(self, index):"""Load image and bounding boxes info from text file in the kaist dataset format."""filename = os.path.join(self._data_path, self._image_set, 'annotations', index + '.txt')# print ('Loading: {}'.format(filename))with open(filename) as f:lines = f.readlines()[1:]num_objs = len(lines)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)seg_areas = np.zeros((num_objs), dtype=np.float32)# Load object bounding boxes into a data frame.ix = 0for obj in lines:# Make pixel indexes 0-basedinfo = obj.split()# jamif self._image_set.find("train") != -1:if info[0] == "person":x1 = float(info[1])y1 = float(info[2])x2 = x1 + float(info[3])y2 = y1 + float(info[4])assert(x2 >= x1)assert(y2 >= y1)cls = self._class_to_ind['pedestrian']boxes[ix, :] = [max(x1 - 1, 0), max(y1 - 1, 0), min(x2 - 1, 639), min(y2 - 1, 479)]gt_classes[ix] = clsoverlaps[ix, cls] = 1.0seg_areas[ix] = (x2 - x1 + 1) * (y2 - y1 + 1)ix = ix + 1overlaps = scipy.sparse.csr_matrix(overlaps)return {'boxes': boxes,'gt_classes': gt_classes,'gt_overlaps': overlaps,'flipped': False,'seg_areas': seg_areas}

所以，现在修改_load_revised_annotation(self, index)函数，下面是修改之后的，（就简单的把上述的person,person?people cyclist都认为是pedestrian）

    def _load_revised_annotation(self, index):"""Load image and bounding boxes info from text file in the kaist dataset format."""filename = os.path.join(self._data_path, self._image_set, 'annotations', index + '.txt')# print ('Loading: {}'.format(filename))with open(filename) as f:lines = f.readlines()[1:]num_objs = len(lines)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)seg_areas = np.zeros((num_objs), dtype=np.float32)# Load object bounding boxes into a data frame.ix = 0for obj in lines:# Make pixel indexes 0-basedinfo = obj.split()# jam#if self._image_set.find("train") != -1:#    if info[0] == "person":x1 = float(info[1])y1 = float(info[2])x2 = x1 + float(info[3])y2 = y1 + float(info[4])assert(x2 >= x1)assert(y2 >= y1)cls = self._class_to_ind['pedestrian']boxes[ix, :] = [max(x1 - 1, 0), max(y1 - 1, 0), min(x2 - 1, 639), min(y2 - 1, 479)]gt_classes[ix] = clsoverlaps[ix, cls] = 1.0seg_areas[ix] = (x2 - x1 + 1) * (y2 - y1 + 1)ix = ix + 1overlaps = scipy.sparse.csr_matrix(overlaps)return {'boxes': boxes,'gt_classes': gt_classes,'gt_overlaps': overlaps,'flipped': False,'seg_areas': seg_areas}

注意，要删除原来生成的 cache里的文件

否则还是和之前一样。
这样就终于训练成功了。