How do I create a labeling task and assign data to it through the CVAT API?

Question

I'm trying to use the CVAT API to create labeling tasks and attach image data from AWS S3 to the tasks. However, I find it impossible to assign segments to a labeling task while creating the task or after creating the task. When I use the UI to create tasks, they are subsequently shown in the "Tasks" section. When I use API requests, they are somehow registered, but not shown in the UI, and no data seems to be attached to them. I have tried various routes from the CVAT OpenAPI docs, but none of them seem to offer the capabilities that I need.

Answer 1

I was able to do it in the end by inspecting the network traffic in the browser and copying the POST-requests made when creating a task.

Answer 2

Per https://github.com/opencv/cvat/issues/6012, here is code that creates a CVAT task from S3:

import time
from http import HTTPStatus

from cvat_sdk import make_client
from cvat_sdk.api_client import ApiClient, Configuration
from cvat_sdk.core.proxies.tasks import ResourceType


def low_level_client(ip, user, password):
    host = f"http://{ip}:8080"
    return ApiClient(Configuration(host, username=user, password=password))


def create_task_low_level(api_client, spec, data):
    task, task_response = api_client.tasks_api.create(spec)
    assert task_response.status == HTTPStatus.CREATED
    api_client.tasks_api.create_data(task.id, data)
    for i in range(10):
        status = api_client.tasks_api.retrieve_status(task.id)[0]
        if status.state.value == "Finished":
            break
        time.sleep(1)
    return task


ip = "1.2.3.4"
manifest = "da787e7156274df9a4fcc63e48acb28f.jsonl" 
filenames = [
    "OK01R_left_2023-03-31T08:27:49.501349+00:00.jpg",
    "OK01R_right_2023-03-31T08:27:50.358855+00:00.jpg",
]
images_and_manifest = filenames + [manifest]
storage_id = 4                 
data = dict(
    image_quality=75,
    use_cache=True,
    cloud_storage_id=storage_id,
    server_files=images_and_manifest,
)
user = "user"
password = "secret"
spec = {"name": "task1", "labels": [{"name": "car"}]}
with low_level_client(ip, user, password) as api_client:
    task = create_task_low_level(api_client, spec, data)