faster-rcnn pkl文件的生成

红太狼 2023-06-21 10:27 8阅读 0赞

参考程序[pascal\_voc.py][pascal_voc.py]中\_load\_pascal\_annotation函数用来读取标注信息，gt\_roidb用于保存或读取pkl文件。

单个文件的标注信息提取流程详见如下程序。

import os
    import scipy
    import numpy as np
    import xml.etree.ElementTree as ET
    
    _classes = ('__background__',
                         'car', 'bus', 'truck')
    num_classes = len(_classes)
    _class_to_ind = dict(list(zip(_classes, list(range(num_classes)))))
    _data_path = 'data/cars' #Annocations的所在目录
    
    def load_pascal_annotation(index):
        """
        Load image and bounding boxes info from XML file in the PASCAL VOC
        format.
        """
        filename = os.path.join(_data_path, 'Annotations', index + '.xml')
        tree = ET.parse(filename)
        objs = tree.findall('object')
        # Load object bounding boxes into a data frame.
        num_objs = len(objs)
        for ix, obj in enumerate(objs):  # xml has empty object
            bbox = obj.find('bndbox')
            if bbox == None:
                num_objs -= 1
                continue
        assert num_objs != 0, '{} has 0 obj'.format(filename)
        tree = ET.parse(filename)
        objs = tree.findall('object')
        size = tree.find('size')
        imgw = float(size.find('width').text)
        imgh = float(size.find('height').text)
        boxes = np.zeros((num_objs, 4), dtype=np.uint16)
        gt_classes = np.zeros((num_objs), dtype=np.int32)
        overlaps = np.zeros((num_objs, num_classes), dtype=np.float32)
        # "Seg" area for pascal is just the box area
        seg_areas = np.zeros((num_objs), dtype=np.float32)
        # Load object bounding boxes into a data frame.
        ix_invalid = 0
        for ix, obj in enumerate(objs):
            cls = _class_to_ind[obj.find('name').text.lower().strip()]  # car_multy_part
            bbox = obj.find('bndbox')
            if bbox == None:
                ix_invalid += 1
                continue
            ix -= ix_invalid
            # print ("bbox: {}, type: {}".format(bbox, type(bbox)))
            # Make pixel indexes 0-based
            x1 = float(bbox.find('xmin').text) - 1
            y1 = float(bbox.find('ymin').text) - 1
            x2 = float(bbox.find('xmax').text) - 1
            y2 = float(bbox.find('ymax').text) - 1
            # assert x2 > x1 and y2 > y1, 'box error {}'.format(index)
            xmin = min(x1, x2)   #标注结果矫正
            xmax = max(x1, x2)
            ymin = min(y1, y2)
            ymax = max(y1, y2)
    
            x1, y1, x2, y2 = max(xmin, 0), max(ymin, 0), min(xmax, imgw), min(ymax, imgh)
            flag_gt = (x2 >= imgw) or (y2 >= imgh)
            flag_invalid = (x2 == 0) or (y2 == 0) or (x1 < 0) or (y1 < 0) or (x1 >= x2) or (y1 >= y2)
            if flag_gt or flag_invalid:
                print('index: {} biaozhuerror'.format(index)) #对于单个物体时，返回错误的gt_boxes等信息
                continue
            # assert (x2 <imgw) and (y2<imgh), 'index: {}'.format(index)
            # cls = self._class_to_ind[obj.find('name').text.lower().strip()] #car_multy_part
            # print ('{:4},box: {:5}, {:5}, {:5}, {:5}'.format(ix, x1, y1, x2, y2))
            boxes[ix, :] = [x1, y1, x2, y2]
            gt_classes[ix] = cls
            overlaps[ix, cls] = 1.0
            seg_areas[ix] = (x2 - x1 + 1) * (y2 - y1 + 1)
        overlaps = scipy.sparse.csr_matrix(overlaps)
        return {'boxes': boxes,
                'gt_classes': gt_classes,
                'gt_overlaps': overlaps,
                'flipped': False,
                'seg_areas': seg_areas}

参考文献：

1.  **[tf-faster-rcnn][]**

[pascal_voc.py]: https://github.com/endernewton/tf-faster-rcnn/blob/master/lib/datasets/pascal_voc.py
[tf-faster-rcnn]: https://github.com/endernewton/tf-faster-rcnn