pascal_voc_tools package

Subpackages

• pascal_voc_tools.Augmenter package
  • Submodules
  • pascal_voc_tools.Augmenter.bbox_util module
  • pascal_voc_tools.Augmenter.data_aug module
  • Module contents
• pascal_voc_tools.Evaluater package
  • Submodules
  • pascal_voc_tools.Evaluater.tools module
  • Module contents
• pascal_voc_tools.utils package
  • Submodules
  • pascal_voc_tools.utils.nms module
  • Module contents

Submodules

pascal_voc_tools.anchors_kmeans module

Calculating anchors using k-means for yolov3.

class pascal_voc_tools.anchors_kmeans.AnchorsKMeans(filelist)

Bases: object

对anchors聚类

calculate(num_clusters, yolo_input_shape, yolo_version='yolov3')

Parameters:

• num_clusters – number of clusters;
• yolo_version – default=’yolov3’, yolov2 or yolov3;
• yolo_input_shape – input images shape，multiples of 32. etc. 416*416;

Returns:

anchors

save(output_dir)

Save current anchors to a file

Parameters:output_dir – str, a dir to save anchors file.

pascal_voc_tools.annotation_tools module

pascal_voc_tools.aspect_ratios module

pascal_voc_tools.darknet_config module

class pascal_voc_tools.darknet_config.DarknetConfig

Bases: object

Tools about config file for darknet.

change_anchors(layers, new_anchors)

Parameters:

• layers – list, the list of Layers;
• new_anchors – list, which have n anchors, for each anchor it has the format like [w, h].

Returns:

list, the layers had been changed.

Return type:

layers

check_yolo(cfg_file, label_list_file)

Using to check some key in yolo config. :param cfg_file: str, the config file path; :param label_list_file: str, the file of label list;

parse(config_path)

decode the config file for darknet :param config_path: str, the path of config file;

Returns:list, including Layer struct. Return type:layers

save(layers, save_path)

Saving the list of Layer to save_path. :param layers: list, have some Layers; :param save_path: str, the path using to save.

class pascal_voc_tools.darknet_config.Layer(name)

Bases: object

Darknet Layer Struct

pascal_voc_tools.darknet_config.main()

pascal_voc_tools.datatools module

pascal_voc_tools.file module

pascal_voc_tools.format module

pascal_voc_tools.image_annotation_split module

pascal_voc_tools.resize module

pascal_voc_tools.tools module

Some useful tools.

pascal_voc_tools.tools.average_precision_11_point()

pascal_voc_tools.tools.average_precision_matric()

pascal_voc_tools.tools.bb_intersection_over_union(box_a, box_b)

calculate intersection over union between two boundboxes.

Parameters:

• box_a – list of xmin, ymin, xmax, ymax;
• box_b – list of xmin, ymin, xmax, ymax;

Returns:

a float number of iou between two inputs.

Module contents

Pascal Voc Tools

This package provide some tools using for pascal voc format dataset and some usrful functions.

class pascal_voc_tools.ImageWrapper

Bases: object

Some image tools.

crop_image(split_bboxes)

Split an image to some subimages.

Parameters:split_bboxes – list, like[[xmin, ymin, xmax, ymax], ] Returns:list, all subimages. Return type:images

load(image_path: str)

load image data.

Parameters:image_path – image file path.

resize_by_rate(rate: float)

Resize image.

Parameters:rate – image resize rate.

resize_letter_box(width: int, height: int)

using letter box to resize image data.

Parameters:

• width – new image width.
• height – new image height.

Returns:

a flot of resize rate.

save(save_path)

save image data to a file.

Parameters:save_path – a image file path.

class pascal_voc_tools.DataSource(database: str = 'Unknown')

Bases: object

Source data format in Pascal VOC xml

database

a str of dataset name.

load_from_node(node)

Decode database info from node.

Parameters:node – a source Element node.

save_to_node(node)

Encode database info from node.

Parameters:node – a source Element node.

class pascal_voc_tools.ImageSize(width: int = 0, height: int = 0, depth: int = 0)

Bases: object

Size data format in Pascal VOC xml

width

a int of image width.

height

a int of image height.

depth

a int of image depth.

load_from_node(node)

Decode size info from node.

Parameters:node – a size Element node.

save_to_node(node)

Encode size info to node.

Parameters:node – a size Element node.

class pascal_voc_tools.PascalXml(folder: str = '', filename: str = '', path: str = '', source: pascal_voc_tools.xml_tools.DataSource = <pascal_voc_tools.xml_tools.DataSource object>, size: pascal_voc_tools.xml_tools.ImageSize = <pascal_voc_tools.xml_tools.ImageSize object>, segmented: int = 0, object_list: list = None)

Bases: object

Pascal VOC xml file data format

folder

image folder.

filename

image file name.

path

image file path.

source

the DataSource.

size

the ImageSize.

segmented

default 0.

object_list

a list of XmlObject

add_object(name, xmin, ymin, xmax, ymax, pose='Unspecified', truncated=0, difficult=0)

add an object info

Parameters:

• name – str, class name.
• xmin – int, left.
• ymin – int, top.
• xmax – int, right.
• ymax – int, bottom.
• pose – str, default is ‘Unspecified’.
• truncated – str, default is 0.
• difficult – int, default is 0.

convert2csv(classes: list = [])

Convert data to info list. The head is [file_name, width, height, category, xmin, ymin, xmax, ymax]

Parameters:classes – a list of categories.

convert2yolotxt(save_path: str, classes: list)

save data to txt for yolo format.

Parameters:

• save_path – the txt file path to save data.
• classes – a list of categories.

crop_annotations(split_bboxes, iou_thresh=0.7)

Using split_bboxes to split an xml file.

Parameters:

• xml_info – dict, all info about a xml.
• split_bboxes – list, like [[xmin, ymin, xmax, ymax], ]

Returns:

list, like [xml_info, ]

Return type:

subannotations

load(xml_file_path)

form a xml file load data.

Parameters:xml_file_path – a xml file path.

load_from_node(node)

load data from Element

Parameters:node – an annotation Element.

replace_name(old_name, new_name)

Replace an object name.

Parameters:

• old_name – str, an object class name.
• new_name – str, a new object class name.

resize_obj_by_rate(rate: float, biases: tuple)

Resize all bndbox by rate.

Parameters:rate – a float for timeing bndbox.

save(save_xml_path)

save data to a xml file.

Parameters:save_xml_path – the xml path to save data.

save_to_node(node)

save info to annotation Element

Parameters:node – an annotation Element.

class pascal_voc_tools.JPEGImages(image_dir=None)

Bases: object

Some tools for JPEGImages dir.

load(image_dir=None)

Load jpg image files’ path.

Parameters:image_dir – a directory incluing jpg images.

class pascal_voc_tools.Annotations(ann_dir)

Bases: object

deal with annotations in PascalVOC format dataset.

ann_dir

str, the dir including xmls.

Raises:AssertError – can not find path.

convert2csv(save_path, category_list=[])

find all bndbox info and save to csv file. If category_list exist, only save the categories in list.

Parameters:

• save_path – a csv file path.
• category_list – a list for keep category to save.

Raises:

AssertionError – the save path not a csv file path.

count_category()

get_bbox_info()

height_analyse(save_dir='./', split_num=100)

Draw height ratio distribution in dataset.

Parameters:

• save_dir – str, image save dir;
• split_num – int, the number of data partitions

iou_analyse(save_dir='./', split_num=100)

Draw iou distribution in dataset.

Parameters:

• save_dir – str, image save dir;
• split_num – int, the number of data partitions

load(name_list=None)

width_analyse(save_dir='./', split_num=100)

Draw width ratio distribution in dataset.

Parameters:

• save_dir – str, image save dir;
• split_num – int, the number of data partitions

class pascal_voc_tools.VOCTools(voc_root_dir: str)

Bases: object

a VOC format dataset.

root_dir

a voc directory like VOC2007.

check_jpg_xml_match()

Check matching degree about xml files and jpeg files.

crop_data(save_root_dir, set_name_list, min_side=1000, max_side=1333, cover_thresh=0.2, iou_thresh=0.7)

split image and annotation to some sub data.

Parameters:

• save_root_dir – a new dir to save new data.
• set_name_list – a list like [train, val].
• min_side – a int of sub image min side.
• max_side – a int of sub image max side.
• cover_thresh – sub image overlap rate.
• iou_thresh – bndbox with sub image overlap rate.

Returns:

a new VOCTools including new data.

crop_image_annotations(jpg_path, xml_path, min_side, max_side, cover_thresh=0.2, iou_thresh=0.7)

Split an image and it’s annotation file.

Parameters:

• jpg_path – str, image path.
• xml_path – str, xml path.
• cover_thresh – float, cover rate about each subimage, default=0.2.
• iou_thresh – float, filter annotations which one’s iou is smaller than thresh, default=1.0.

Returns:

a list of ImageWrapper. subannotations: a list of PascalXml

Return type:

subimages

gen_format_dir()

get_crop_bboxes(width, height, min_side, max_side, cover_thresh=0.2)

get_image_path_by_xml_path(xml_path)

using xml path to inference image path

Parameters:xml_path – str, xml file path. Returns:image path in current dataset.

get_year()

match_jpg_xml()

resize_by_size(width, height, save_root_dir)

Resize whole dataset by set a fix image size.

Parameters:

• width – int, new image width.
• height – int, new image height.
• save_root_dir – a new dir to save data.

Returns:

new VOCTools have new data.

class pascal_voc_tools.AnchorsKMeans(filelist)

Bases: object

对anchors聚类

calculate(num_clusters, yolo_input_shape, yolo_version='yolov3')

Parameters:

• num_clusters – number of clusters;
• yolo_version – default=’yolov3’, yolov2 or yolov3;
• yolo_input_shape – input images shape，multiples of 32. etc. 416*416;

Returns:

anchors

save(output_dir)

Save current anchors to a file

Parameters:output_dir – str, a dir to save anchors file.

class pascal_voc_tools.DarknetConfig

Bases: object

Tools about config file for darknet.

change_anchors(layers, new_anchors)

Parameters:

• layers – list, the list of Layers;
• new_anchors – list, which have n anchors, for each anchor it has the format like [w, h].

Returns:

list, the layers had been changed.

Return type:

layers

check_yolo(cfg_file, label_list_file)

Using to check some key in yolo config. :param cfg_file: str, the config file path; :param label_list_file: str, the file of label list;

parse(config_path)

decode the config file for darknet :param config_path: str, the path of config file;

Returns:list, including Layer struct. Return type:layers

save(layers, save_path)

Saving the list of Layer to save_path. :param layers: list, have some Layers; :param save_path: str, the path using to save.