API

download

Function Description :

Download SVHN dataset from the original svhn dataset

Optional features :

Extract the downloaded dataset.
Delete the .tar.gz file after downloading.

Parameter Description
Parameter	Default Value	Description
dataset_type	‘train’	Dataset type. One of ‘train’, ‘test’ or ‘extra’
save_path	‘’	Dataset download path / extraction path. The path should not contains any trailing ‘/’
extract	True	Whether or not the downloaded .tar.gz file should be extracted
force	False	Download and save the dataset even if it’s already in the given directory
del_zip	False	Whether or not the .tar.gz file should be deleted after extraction

Output view :

../data/svhn/train : params : (‘train’, ‘../data/svhn’, extract=True)

../data/svhn/train.tar.gz : params : (‘train’, ‘../data/svhn’, extract=False)

ann_to_csv

Function Description :

Convert .mat file to .json file

Optional features :

Set bbox format to KITTI / Normalilzed

Parameter Description
Parameter	Default value	Description
file_path	None	.mat file path. Ex: relative path ‘../data/digitStruct.mat’ or absolute path ‘C:/usr/local/data/digitStruct.mat’
save_path	None	.json file path Ex : ‘../data/train.json’
bbox_type	‘normalize’	Two types of bounding box declaration formats: ‘normalize’ or ‘kitti’

Output view :

[
   {
      "boxes": [
      {
         "width": 81.0,
         "top": 77.0,
         "label": 1.0,
         "left": 246.0,
         "height": 219.0
      },
      {
         "width": 96.0,
         "top": 81.0,
         "label": 9.0,
         "left": 346.0,
         "height": 119.0
      }
      ],
      "filename": "1.png"
   },
   ...
]

ann_to_json

Function Description :

Convert .mat file to .csv file

Optional features :

Set bbox format to KITTI / Normalilzed

Parameter Description
Parameter	Default value	Description
file_path	None	.mat file path. Ex: relative path ‘../data/digitStruct.mat’ or absolute path ‘C:/usr/local/data/digitStruct.mat’
save_path	None	.json file path Ex : ‘../data/train.csv’
bbox_type	‘normalize’	Two types of bounding box declaration formats: ‘normalize’ or ‘kitti’

Output view :

train.csv
filename	class	left	top	width	height
1.png	1	81.0	77.0	246.0	219.0
1.png	9	96.0	81.0	346.0	119.0

gen_dataset

Function Description :

Generate the image dataset and associated, readily-accessible annotation file.

Optional features :

Convert the RGB image to Gray-scale
Set the threshold for the minimum # of digits present
Set the threshold for the maximum # of digits present
Crop the image to only include the digits with minimal background
Resize the images
Generate the annonations in two ways; In the first way the resulting annotations are directly compatible with the fixed numbered MDR task. The second annotation type is in a more generalized form. The resulting .json file could be converted into any well known annotation type [PascalVOC, COCO, YOLO-darkNet]. For further image annotation conversions, you could use the python library called imgann

Parameter Description
Parameter	Default Value	Description
image_path	None	The image folder path (not the .tar.gz path) without trailing ‘/’. Ex : ‘../data/train’
mat_path	None	.mat file path Ex : ‘../data/train/digitStruct.mat’
rgb	True (bool)	Whether to convert to RGB format
min_digits	0	The minimum number of digits in the SVHN image
max_digits	6	The maximum number of digits in the SVHN image inclusive value
crop	True	Whether to crop the digit containing part from the original image
resize_shape	(64, 64)	Shape to resize the image to. Could be rectangular or square in shape
only_labels	False	If true, outputs only the labels in a numpy.ndarray. Else, outputs the formal json annotation file
save	False	Whether or not to save the returned files

Output view :

{
   "annotations": [
      {
            "id": "1",
            "image_id": "1",
            "category_id": 1,
            "area": 22165,
            "bbox": [170, 114, 313, 269],
            "ignore": "0",
            "iscrowd": "0"
      },
      .
      .
   ],
   "images": [
      {
            "file_name": "1.jpg",
            "height": 413,
            "width": 413,
            "id": "1"
      },
      .
      .
   ],
   "categories": [
      {
            "id": 1,
            "name": 1,
            "supercategory": "none"
      },
      .
      .
   ],
}