DataGym.ai
Search…
Project
Dataset
Label configuration
Label mode
AI-Assistant
API Token
Account-Management
Import Data
How to import labels
DataGym can import your labeled data. How this works and the requirements for this task are described in this document.
The requirements:
- The project id is also required. This id can be found within the URL of the project's detail page or can be found with the API. Try the endpoint to list all Projects.
- In addition, the internal ids of all affected images are required. They can be caught via the API. Try the endpoint to get a dataset by its id.
To use the import feature the import data must match the label configuration.
The import:
You can import the labeled data either via our python package using the public api or upload the JSON formatted file from the project overview page.
Import & export labeled data is available within the project home page.
The import data format is similar to the export format and described in the following sections.
Example label configuration:
As an example to visualize the import, the following configuration is used. It's the same configuration used in the Export Data section of this documentation:
Within the exported data the label_classes look like the following snippet from the export data section:
label_classes.json
1
{
2
"label_classes": [
3
{
4
"class_name": "bus",
5
"geometry_type": "rectangle"
6
},
7
{
8
"class_name": "car",
9
"geometry_type": "polygon"
10
},
11
{
12
"class_name": "car_type",
13
"classification_type": "select"
14
},
15
{
16
"class_name": "weather",
17
"classification_type": "radio"
18
},
19
{
20
"class_name": "daytime",
21
"classification_type": "select"
22
}
23
]
24
}
Copied!
The used class names are required for the import. Also, the used geometry type or classification type defines the import format.
Classifications of type option are exported as type radio if they contain up to three options otherwise they are exported as type select.
The section "label_classes" is not part of the import JSON, but useful to understand how the import works.
Lets look into the expected data:
The import format:
In its simplest form, the import looks like this:
1
[
2
{
3
"internal_media_ID": "<media_id>",
4
"keepData": false,
5
"global_classifications": {},
6
"labels": {}
7
}
8
]
Copied!
Property
Description
internal_media_ID
The internal id to identify the media.
keepData
If keepData is equal to false, all already existing labels for the current Image will be deleted after the labels upload.
If keepData is equal to true, all new labels will be added to the already existing labels for the current Image.
Default value for keepData is true
global_classifications
Uses the label_classes to describe the image.
labels
Uses the class names to describe a geometry within the image.
To do a successful import a valid media_id in the internal_media_ID node is required.
Either of the other three properties is optional.
Schema validation
To see if your import Json is valid, you can validate it against this JSON Schema.
1
{
2
"$schema": "http://json-schema.org/draft-07/schema#",
3
"title": "Json Import Schema",
4
"description": "Validate the user input",
5
"definitions": {
6
"rectangle": {
7
"type": "array",
8
"items": {
9
"$ref": "#/definitions/point_def_rectangle"
10
},
11
"minItems": 1,
12
"maxItems": 1
13
},
14
"polygon": {
15
"type": "array",
16
"items": {
17
"$ref": "#/definitions/point_def_not_rectangle"
18
},
19
"minItems": 3
20
},
21
"line": {
22
"type": "array",
23
"items": {
24
"$ref": "#/definitions/point_def_not_rectangle"
25
},
26
"minItems": 2
27
},
28
"point": {
29
"type": "array",
30
"items": {
31
"$ref": "#/definitions/point_def_not_rectangle"
32
},
33
"minItems": 1,
34
"maxItems": 1
35
},
36
"numberOrString": {
37
"oneOf": [
38
{
39
"type": "number",
40
"minimum": 0
41
},
42
{
43
"type": "string",
44
"pattern": "^[0-9]+quot;
45
}
46
]
47
},
48
"point_def_not_rectangle": {
49
"type": "object",
50
"properties": {
51
"x": {
52
"$ref": "#/definitions/numberOrString"
53
},
54
"y": {
55
"$ref": "#/definitions/numberOrString"
56
}
57
},
58
"minProperties": 2,
59
"maxProperties": 2,
60
"required": [
61
"x",
62
"y"
63
]
64
},
65
"point_def_rectangle": {
66
"type": "object",
67
"properties": {
68
"x": {
69
"$ref": "#/definitions/numberOrString"
70
},
71
"y": {
72
"$ref": "#/definitions/numberOrString"
73
},
74
"w": {
75
"$ref": "#/definitions/numberOrString"
76
},
77
"h": {
78
"$ref": "#/definitions/numberOrString"
79
}
80
},
81
"required": [
82
"x",
83
"y",
84
"w",
85
"h"
86
]
87
},
88
"classification": {
89
"type": "array",
90
"items": [
91
{
92
"oneOf": [
93
{
94
"type": "string"
95
},
96
{
97
"type": "null"
98
}
99
]
100
},
101
{
102
"oneOf": [
103
{
104
"type": "string"
105
},
106
{
107
"type": "object",
108
"patternProperties": {
109
"": {
110
"$ref": "#/definitions/classification"
111
}
112
}
113
}
114
]
115
}
116
],
117
"additionalItems": {
118
"type": "string"
119
}
120
},
121
"geometry": {
122
"type": "array",
123
"items": {
124
"oneOf": [
125
{
126
"$ref": "#/definitions/point_def_not_rectangle"
127
},
128
{
129
"$ref": "#/definitions/point_def_rectangle"
130
}
131
]
132
}
133
}
134
},
135
"type": "array",
136
"minItems": 1,
137
"items": {
138
"type": "object",
139
"properties": {
140
"internal_media_ID": {
141
"type": "string"
142
},
143
"keepData": {
144
"type": "boolean"
145
},
146
"global_classifications": {
147
"type": "object",
148
"patternProperties": {
149
"": {
150
"$ref": "#/definitions/classification"
151
}
152
}
153
},
154
"labels": {
155
"type": "object",
156
"patternProperties": {
157
"": {
158
"type": "array",
159
"items": {
160
"type": "object",
161
"properties": {
162
"geometry": {
163
"$ref": "#/definitions/geometry"
164
},
165
"rectangle": {
166
"$ref": "#/definitions/rectangle"
167
},
168
"polygon": {
169
"$ref": "#/definitions/polygon"
170
},
171
"point": {
172
"$ref": "#/definitions/point"
173
},
174
"line": {
175
"$ref": "#/definitions/line"
176
},
177
"image_segmentation": {
178
"type": "array"
179
},
180
"classifications": {
181
"type": "object",
182
"patternProperties": {
183
"": {
184
"$ref": "#/definitions/classification"
185
}
186
}
187
},
188
"nested_geometries": {
189
"patternProperties": {
190
"": {
191
"type": "array",
192
"items": {
193
"type": "object",
194
"properties": {
195
"geometry": {
196
"$ref": "#/definitions/geometry"
197
},
198
"rectangle": {
199
"$ref": "#/definitions/rectangle"
200
},
201
"polygon": {
202
"$ref": "#/definitions/polygon"
203
},
204
"point": {
205
"$ref": "#/definitions/point"
206
},
207
"line": {
208
"$ref": "#/definitions/line"
209
},
210
"classifications": {
211
"type": "object",
212
"patternProperties": {
213
"": {
214
"$ref": "#/definitions/classification"
215
}
216
}
217
},
218
"additionalProperties": false
219
}
220
}
221
}
222
}
223
}
224
},
225
"additionalProperties": false
226
}
227
}
228
}
229
}
230
}
231
}
232
}
233
​
Copied!
For validation you can either use one of the validators from the official JSON Schema page. Those are available for several programming languages and the console. Or you can use one of many JSON Schema online validators.
Properties explained
Lets have a look into the global_classifications property:
This property may look like the following snippet:
global_classifications.json
1
{
2
"global_classifications": {
3
"weather": [
4
"sunny",
5
{
6
"daytime": [
7
"day"
8
]
9
}
10
]
11
}
12
}
Copied!
This example can also be found within the Export data page. Nested classifications are possible. If a value is not defined use null.
The possible object keys and their values are defined in the label configuration.
Lets have a look into the labels object:
The snippet may look like the following:
labels_snippet.json
1
{
2
"labels" : {
3
"bus": [],
4
"car": []
5
}
6
}
Copied!
This object holds all classes of geometries defined within the label configuration and found within your image. Each entry is a list of geometries. Each geometry in such a list consists of two entries:
Property
Description
geometry
This section defines the area of interest within the image. The structure depends on the type of geometry. (Deprecated, use the appropriate entry from one of the properties below)
polygon
Use this property if you want to import a polygon
line
Use this property if you want to import a line
point
Use this property if you want to import a point
rectangle
Use this property if you want to import a rectangle
classifications
This section is similar to the global_classifications section described above.
nested_geometries
This section holds an array of individual geometries (like those above)
image_segmentation
You cant import image segmentation values yet
Let's define a car to look at:
car.json
1
{
2
"car": [
3
{
4
"polygon": [
5
{
6
"x": 57,
7
"y": 412
8
},
9
{
10
"x": 54,
11
"y": 463
12
},
13
{
14
"x": 130,
15
"y": 524
16
}
17
],
18
"classifications": {
19
"car_type": [
20
null
21
]
22
}
23
}
24
]
25
}
Copied!
The car's geometry type is defined as a polygon. See the label_classes.json or the label configuration as reference. To define a polygon a list of points is required. You need to enter at least three points for a valid polygon, but you can enter as many as needed.
line.json
1
{
2
"street": [
3
{
4
"line": [
5
{
6
"x": 57,
7
"y": 412
8
},
9
{
10
"x": 54,
11
"y": 463
12
}
13
],
14
"classifications": {}
15
}
16
]
17
}
Copied!
This is an example for a line. Lines are defined as polylines, therefore they need to have at least two points but can have as many as needed.
point.json
1
{
2
"poi": [
3
{
4
"point": [
5
{
6
"x": 57,
7
"y": 412
8
}
9
],
10
"classifications": {}
11
}
12
]
13
}
Copied!
A point geometry may only have one point containing x and y values.
A bus may look like the following snippet:
bus.json
1
{
2
"bus": [
3
{
4
"rectangle": [
5
{
6
"x": 272,
7
"y": 302,
8
"w": 324,
9
"h": 169
10
}
11
],
12
"classifications": {}
13
}
14
]
15
}
Copied!
The geometry properties now define a rectangle with the start coordinates x & y and the properties w & h for its width and height. In addition, the classifications object is empty because bus has no classifications.
Last modified 1yr ago
Copy link