mindkosh 1.1.4

Mindkosh is the platform for labeling multi-sensor data, from Lidar point clouds to multiple camera images. Our powerful automatic annotation features and a particular focus on making Quality checking easier and faster, helps ML teams obtain high quality labeled datasets at scale.

Read documentation here

Table of contents:

• Table of contents:
• Setup
  • Requirements
  • Installation
• Getting started
• Datasets
  • Create dataset
  • Get datasets
  • Get dataset files
  • Update tags
  • Upload data
  • Upload pointcloud data
  • Upload imagefiles
  • Delete files from dataset
  • Delete dataset
• Projects
  • Create project
  • Get all projects
  • Update project details
  • Delete project
• Tasks
  • Create task
  • Get all tasks
  • Update task details
  • Download annotations
  • Upload annotations
  • Delete task
• Frames
  • Get frames
  • Download frame
  • Show frame annotations
  • Download frame annotations
  • Visualize frame annotations

Setup

Requirements

• Python >= 3.7
• Account on the Mindkosh platform (Create on now)
• SDK Token (contact us to get one)

Installation

• Install the package

pip install mindkosh

Getting started

from mindkosh import Client, PointCloudFile, ImageFile, Label
client = Client(token)

To get started, create a Task/Project object using the token provided to you. You can also set the token in an environment variable(MK_TOKEN) instead of passing as a parameter to the object constructor.

Datasets

Create dataset

client.create_dataset(
    name: str, 
    data_type: str
) 

Parameters:

• name - unique dataset name
• data_type - image, pointcloud or video

Example:

dataset = client.create_dataset(
    name = 'test-1',
    data_type = 'pointcloud'
)

Get datasets

datasets = client.get_datasets(
    dataset_id = 1
)

Get dataset files

files = client.get_dataset_files(
    dataset_id = 1,
    max_files = 100
)

Update tags

updated_dataset_files = client.update_tags(
    dataset_id = 1,
    datasetfile_ids = [1,2,3,4,5,6,8],
    add = ['tag1','tag2'],
    remove = ['tag2'],
    all = False
)

Upload data

Uploads images from a list of directories or/and files. Useful when all the images same tags/extra.

client.upload_data(
    dataset_id = 1,
    tags = ['tag4'],
    resources = ['/home/user/Desktop/my-images/', '/home/user/Downloads/test1.jpg'],
    recursive = False,
    manifest_file = None,
    save_files_dir = None,
    **kwargs
)

Upload pointcloud data

Uploads images from a list of directories or/and files. Useful when all the images same tags/extra.

pcdfile1 = PointCloudFile(
    filepath = '/file/path1/',
    related_files = [
        ImageFile(
            filepath='/path/to/image1',
            tags=[],
            extra={
                "intrinsic": [255.520403, 449.883682, 250.828738, 255.237477],

                # Projection matrix from lidar to camera
                "extrinsic": [
                    [-0.735827, -0.65789, -0.0384775, 0],
                    [0.6567956, -0.70452916, 0.00140833, 0],
                    [-0.02255652, -0.0248216, 0.9993586, 0],
                    [0, 0, 0, 1]
                ],

                # Camera projection model - PINHOLE OR FISHEYE
                # For FISHEYE, mirrorParameter is also needed
                "cameraModel": "PINHOLE",
                "device_id": 1
            }
        )
    ]
)

client.upload_pointcloud_data(
    dataset_id = 1,
    pcdfiles = [pcdfile1, pcdfile2]
    **kwargs
)

Upload imagefiles

Uploads images with tags/extra for each image

client.upload_imagefiles(
    dataset_id = 1,
    imagefiles = [
        ImageFile(
            filepath='/path/to/image1',
            tags=["city1"]
        ),
        ImageFile(
            filepath='path/to/image2',
            tags=["city2"]
        )
    ]
)

Delete files from dataset

Delete select files or delete all the files from a dataset

client.delete_files_from_dataset(
    dataset_id = 1,
    file_ids = [1,2,3,4],
    delete_all = False
)

Delete dataset

Delete a dataset along with its files

client.delete_dataset(dataset_id=1)

Projects

Create project

project = client.project.create(name, description=None) 

To create a project, enter a name and an optional description.

Get all projects

Get list of all project objects

myprojects = client.project.get()

Update project details

project.update_name("new name")
project.update_description("New description")

Delete project

project.delete()

Tasks

Create task

Create a new task :

labels = [
    Label(
      name='car',
      color='#fffccc',
      sequence=1,
      type='non_mask'
    ),
    Label(
      name='bus',
      color='#fcf0fc',
      sequence=2,
      type='instance_mask'
      attributes =[]
    )
]

Note : Each label should have a unique name.

task = client.task.create( 
    name='task1',
    labels=labels,
    dataset_id=1,
    tags=['tag1'],
    project_id=None,
    job_modes=['validation','qc']
    batches=3
)

To create a task and upload images/video, user must provide either resources/manifest.

Parameters:

• name - A valid task name.
• batches - Number of images in each job. Only valid for resource_type="imageset".
• project_id - Optional - ID of the project this task should belong to.

Get all tasks

Get list of all the tasks

mytasks = client.task.get()

Update task details

task.update_name("New name")
task.update_project_id(new_valid_project_id)  

Download annotations

Get releases

releases = task.get_releases()

Create a release

releases = task.create_release(
    format="coco", 
    batches=[1,2], 
    description=None, 
    webhook_url=None
)

format can be any of the following:

• Available formats and their parameter names:
• COCO - "coco"
• Datumaro - "datumaro"
• Pascal VOC - "voc"
• Segmentation mask - "segmentation_mask"
• YOLO - "yolo"

Download a release

task.download_release(
    release_id=1,
    local_path='/local/dir/'
)

Delete a release

task.delete_release(release_id=1)

Upload annotations

task.upload_annotations(
    annotation_format='coco',
    local_path='annotation_file.zip'
) 

Delete task

Delete a task -

task.delete()

Frames

Get frames

frames = t1.frames()
frame = frames[frame_id]

Download frame

frame.download(location)

Show frame annotations

frame.annotations

Download frame annotations

frame.download_annotations(location,format=None)

Parameters :

• location - local dir location where you want to save annotations
• format - None if you want to download row annotations. Any of ("coco", "datumaro", "voc", "yolo") if you want to download datasets in specific format.

Visualize frame annotations

frame.visualize(show_annotations=True,fill_color=0.30)

Parameters :

• show_annotations - False to see raw image, True to see image with annotations.
• fill_color - 0 <= Color transparency <= 1