Task creation
You can add tasks to your batches. It works for any kind of project. For Data processing projects, your can attach a dataset to your batch, add files on it and refer to those files when you create your tasks.
Code Reference
See the code reference for further details.
Usage for annotation tool projects
Add simple task for image
from isahitlab.client import IsahitLab
lab = IsahitLab()
tasks_input = [
{
"name": "folder/01.png",
"resources": [
"/path/to/folder/01.png"
]
}
]
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
tasks=tasks_input
)
Resources
Resources can be a path to a local file or a public url. The local files will be uploaded and stored by Isahit Lab.
The resources field is a list of image to support "sequence" annotation.
If your project is not configured for image sequence annotation, it will only use the first resource.
Add task with annotations
Geometry formats
See Geometry formats to know more about the different formats for rectangles, polygons etc...
from isahitlab.client import IsahitLab
lab = IsahitLab()
tasks_input = [
{
"name": "my_task_01.png",
"resources": [
"/path/to/folder/01.png"
],
"data": {
"annotations": [
{
"polygons": [
{
"geometry": {
"vertices": [
0.05958132045088567,
0.2894211576846307,
0.41867954911433175,
0.7305389221556886
],
"type": "rectangle"
}
}
]
}
]
}
}
]
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
tasks=tasks_input
)
Add task with annotations and labels
from isahitlab.client import IsahitLab
lab = IsahitLab()
tasks_input = [
{
"name": "my_task_01.png",
"resources": [
"/path/to/folder/01.png"
],
"data": {
"annotations": [
{
"polygons": [
{
"geometry": {
"vertices": [
0.05958132045088567,
0.2894211576846307,
0.41867954911433175,
0.7305389221556886
],
"type": "rectangle"
}
}
],
"labels": {
"my_list_id": {
"labels": [
{
"id": "Test",
"name": "Test"
}
]
}
}
}
]
}
}
]
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
tasks=tasks_input
)
Labels
To work properly, my_list_id must match a list id configured on your project
Usage for data processing (form) projects
Data format for Data processing (form) project is totally function of the form configuration of the project.
Resources
Tasks of Data processing (form) project are not resource based so you cannot simply provide the resources field. It will be ignored.
Let's say you have 3 inputs on your form :
| Input ID | Type | Comment |
|---|---|---|
| text-1 | Display text | Image name |
| image-1 | Resource image | Image to be processed |
| listbox-1 | Listbox | For ex. "Is blurry ?" |
from isahitlab.client import IsahitLab
lab = IsahitLab()
tasks_input = [
{
"name": "my_task_01.png",
"data": {
"text-1": "Image my_task_01.png",
"image-1": "https://my-domain.com/public/image.jpg"
}
}
]
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
tasks=tasks_input
)
With resource in dataset
In the example above, we set a public url for image-1. In Isahit Lab, you can create a dataset and attach it to your batch.
Then you can add resources and refer to it in your tasks with resource://folder/image.jpg
Code Reference
See the code reference for further details.
from isahitlab.client import IsahitLab
lab = IsahitLab()
files = [
"image1.jpg",
{"file" : "/path/to/folder/01.png", "path": "folder"}
]
tasks_input = [
{
"name": "image1.jpg",
"data": {
"text-1": "Image image1.jpg",
"image-1": "resource://image1.jpg"
}
}
{
"name": "01.png",
"data": {
"text-1": "Image folder/01.png",
"image-1": "resource://folder/01.png"
}
}
]
lab.append_to_dataset(
project_id='<project_id>',
batch_id='<batch_id>',
files=files)
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
tasks=tasks_input
)
Annotation tool in data processing (form) projects
Your form can contain annotation tools (In this example: iat-1).
from isahitlab.client import IsahitLab
lab = IsahitLab()
files = [
"image1.jpg",
{"file" : "/path/to/folder/01.png", "path": "folder"}
]
tasks_input = [
{
"name": "image1.jpg",
"data": {
"text-1": "Image image1.jpg",
"image-1": "resource://image1.jpg",
"iat-1": {
"resources" : ["resource://image1.jpg"]
}
}
}
{
"name": "01.png",
"data": {
"text-1": "Image folder/01.png",
"image-1": "resource://folder/01.png",
"iat-1": {
"resources" : ["resource://folder/01.png"],
"annotations" : [
... # see "Add task with annotations" above
]
}
}
}
]
lab.append_to_dataset(
project_id='<project_id>',
batch_id='<batch_id>',
files=files)
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
tasks=tasks_input
)
Validation
Unicity
The SDK will automatically raise an error if a task with the same name exists in the batch. The whole import is canceled.
To ignore duplicates and insert others tasks, set raise_if_existing to False
To skip this check and insert duplicates anyway, set disable_unicity_check to True
Data validation
The SDK helps you to import valid data to make sure your tasks will work properly.
If you are encountering problems with validation, you can disable the validation by setting disable_data_check to True
Geometry formats
Rectangle annotation
{
"polygons": [
{
"geometry": {
"vertices": [
0.05958132045088567, # Top Left X relative coordinate
0.2894211576846307, # Top Left Y relative coordinate
0.41867954911433175, # Bottom Right X relative coordinate
0.7305389221556886 # Bottom Right Y relative coordinate
],
"type": "rectangle"
}
}
],
"labels": {
"my_list_id": {
"labels": [
{
"id": "Test",
"name": "Test"
}
]
}
}
}
Polygon, graph and polyline annotation
{
"polygons": [
{
"geometry": {
"vertices": [
0.05958132045088567, # Point X1 relative coordinate
0.2894211576846307, # Point Y1 relative coordinate
0.41867954911433175, # Point X2 relative coordinate
0.7305389221556886, # Point Y2 relative coordinate
...
0.31867954911433175, # Point Xn relative coordinate
0.9305389221556886 # Point Yn relative coordinate
],
"type": "polygon"
}
}
],
"labels": {
"my_list_id": {
"labels": [
{
"id": "Test",
"name": "Test"
}
]
}
}
}
Segmentation
Using helper
For a easier way to import segmentation data, see "Using helper" section below.
In Isahit Lab, the segmentation is "Pixel level" so the native input / output is a png mask.
To work properly, you must provide a base64 encoded mask and annotations list where polygons is a list of three number matching the RGB used in the mask.
Mask
Each labeled pixel in your mask must follow this RGB convention:
| R | G | B |
|---|---|---|
| instance_id % 256 | floor(instance_id / 256) | Label bChannel (in your project configuration) |
newId % 256, Math.floor(newId / 256)
Example
RGB triplet : (1, 0, 7)
Annotations
You must aslo provide an annotation for each RGB triplet in your mask and with the label matching the B channel
{
"polygons": [1, 0, 7],
"labels": {
"my_list_id": {
"labels": [
{
"id": "Test",
"name": "Test"
}
]
}
}
}
Example
from isahitlab.client import IsahitLab
lab = IsahitLab()
tasks_input = [
{
"name": "my_task_01.png",
"resources": [
"/path/to/folder/01.png"
],
"data": {
"annotations": [
{
"polygons": [1, 0, 7]
"labels": {
"my_list_id": {
"labels": [
{
"id": "Test",
"name": "Test"
}
]
}
}
}
],
"mask" : "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAA (truncated)"
}
}
]
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
tasks=tasks_input
)
Using helper
You can use a helper to import segmentation task with Rectangle or Polygon data.
Dependencies
This feature requires additional dependencies. You can install then with pip install isahitlab[image]
from isahitlab.client import IsahitLab
from isahitlab.helpers.labels import extract_labels_map_by_id
from isahitlab.helpers.segmentation import polygon_data_to_segmentation_data
lab = IsahitLab()
tasks_input = [
{
"name": "my_task_01.png",
"resources": [
"/path/to/folder/01.png"
],
"data": {
"annotations": [
{
"polygons": [
{
"geometry": {
"vertices": [
0.05958132045088567,
0.2894211576846307,
0.41867954911433175,
0.7305389221556886
],
"type": "rectangle"
}
}
]
}
]
}
}
]
# Get project configuration and extract label mapping
project_configuration = lab.project_configuration(project_id=project_id)
labels_mapping_by_id = extract_labels_map_by_id(
project_configuration=project_configuration)
# For each task, replace "polygon" or "rectangle" data by "segmentation" data
tasks_input = [{
**t,
"data": polygon_data_to_segmentation_data(
t['data'],
labels_mapping_by_id=labels_mapping_by_id,
image_size={
"width": 800,
"height": 600
}
)
} for t in tasks_input]
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
tasks=tasks_input
)
Image size
image_size parameter is required to build the segmentation mask.
Kili compatibility
The SDK provide a partial compatiblity with Kili format for task creation. Set compatibility_mode to kili.
from isahitlab.client import IsahitLab
lab = IsahitLab()
tasks_input = [
{
"name": "test4.png",
"resources": [
"/home/benjamin/Documents/ISAHIT SAAS/image-demo/image_2/01.png"
],
"data": {
"OBJECT_DETECTION_JOB": {
"annotations": [
{
"boundingPoly": [
{
"normalizedVertices": [
{
"x": 0.09589041095890412,
"y": 0.2607829000381695
},
{
"x": 0.09589041095890412,
"y": 0.16281436871792687
},
{
"x": 0.24738114423851731,
"y": 0.16281436871792687
},
{
"x": 0.24738114423851731,
"y": 0.2607829000381695
}
],
"vertices": [
{
"x": 94.93150684931507,
"y": 163.51087832393227
},
{
"x": 94.93150684931507,
"y": 102.08460918614014
},
{
"x": 244.90733279613215,
"y": 102.08460918614014
},
{
"x": 244.90733279613215,
"y": 163.51087832393227
}
]
}
],
"categories": [
{
"name": "Test1"
}
],
"children": {
"CLASSIFICATION_JOB": {
"categories": [
{
"confidence": 100,
"name": "ENFANT_1"
}
]
},
"CLASSIFICATION_JOB_0": {
"categories": [
{
"children": {
"TRANSCRIPTION_JOB_0": {
"text": "test 2"
}
},
"confidence": 100,
"name": "ENFANT_2"
},
{
"children": {
"TRANSCRIPTION_JOB": {
"text": "test"
}
},
"confidence": 100,
"name": "ENFANT_1"
}
]
}
},
"labelVersion": "default",
"mid": "20241128170655261-1",
"type": "rectangle"
}
]
}
}
}
]
lab.create_tasks(
project_id='<project_id>',
batch_id='<batch_id>',
compatibility_mode="kili",
tasks=tasks_input
)