State-of-the-art Computer Vision with a few lines of code

Report a Bug · Request a Feature . Ask a Question

Ikomia API is an open source tool to easily build and deploy your Computer Vision solutions. You can mix your preferred frameworks such as OpenCV, Detectron2, OpenMMLab or YOLO with the best state-of-the-art algorithms from individual repos.

No effort, just choose what you want and Ikomia downloads it, installs the requirements and runs everything in a few lines of code.

pip install ikomia

Ikomia API has already more than 180 pre-integrated algorithms (mainly OpenCV) but the most interesting algorithms are in Ikomia HUB. That's why, you need to connect to Ikomia HUB when you want to download/install these algorithms.

You can use our demo credentials below or you can get yours for free and join our community here :)

Ikomia authentication is based on two environment variables: IKOMIA_USER=your_login and IKOMIA_PWD=your_password, so you can set these variables by command-lines or use this code snippet:

import os
import ikomia

os.environ['IKOMIA_USER'] = "demo"
os.environ['IKOMIA_PWD'] = "jH4q72DApbRPa4k"

ikomia.authenticate()

When you want to use an algorithm, it's always the same code pattern which is useful when you want to test multiple algorithms effortlessly.

from ikomia.dataprocess import workflow

# Init your workflow
wf = workflow.create("YOLO inference")

# Add YOLO and connect it to your input data
yolo_id, yolo = wf.add_task("infer_yolo_v7")
wf.connect_tasks(wf.getRootID(), yolo_id)

Then run and display your results.

import cv2

# Run directly on your image
wf.run_on(path="path/to/your/image.png")

# YOLO output image with bounding boxes
img_bbox = wf.get_image_with_graphics(yolo_id)
img_bbox = cv2.cvtColor(img_bbox, cv2.COLOR_RGB2BGR)

cv2.imshow("Ikomia Demo", img_bbox)

You can also change each algorithms parameters.

yolo_params = {
    "custom_train": True,
    "custom_model": "path/to/your/model",
    "thr_conf": 0.25
}
wf.set_parameters(task_id=yolo_id, params=yolo_params)

If you don't know what are the parameters (which is often the case), just print your task !

print(yolo)

###################################
#	infer_yolo_v7
###################################

***********************************
*	 PARAMETERS
***********************************

cuda:True
thr_conf:0.25
iou_conf:0.5
pretrain_model:yolov7
custom_model:
img_size:640
custom_train:False

***********************************
*	 INPUTS
***********************************

-----------------------------------
Name: porsche-4795517_960_720
-----------------------------------
Description: 2D or 3D images.
Can be single frame from video or camera stream.
Save folder: 
Auto-save: 0
Data type: image
Save format: .png
Dimension count: 2
File name: 
-----------------------------------
Name: CGraphicsInput
-----------------------------------
Description: Graphics items organized in layer.
Represent shapes and types of objects in image.
Graphics can be created interactively by user.
Save folder: 
Auto-save: 0
Data type: graphics
Save format: .json
Dimension count: 0

***********************************
*	 OUTPUTS
***********************************

-----------------------------------
Name: CImageIO
-----------------------------------
Description: 2D or 3D images.
Can be single frame from video or camera stream.
Save folder: 
Auto-save: 0
Data type: image
Save format: .png
Dimension count: 2
File name: 
-----------------------------------
Name: CObjectDetectionIO
-----------------------------------
Description: Object detection data: label, confidence, box and color.
Save folder: 
Auto-save: 0
Data type: Object detection
Save format: .json
Dimension count: 0

Then you can easily save your workflow as JSON file for reuse.

wf.save("path/to/your/workflow.json")

wf.load("path/to/your/workflow.json")
wf.run_on(path="path/to/your/image.png")

And finally, you can also export your results as JSON files.

# Get all object detection outputs (most of the time, there is only one)
output_list = wf.get_object_detection_output(task_name="infer_yolo_v7")
output_list[0].toJson()

{
  "detections": [
    {
      "box": {
        "height": 195,
        "width": 463,
        "x": 183,
        "y": 261
      },
      "color": {
        "a": 0,
        "b": 192,
        "g": 106,
        "r": 91
      },
      "confidence": 0.90673828125,
      "id": 0,
      "label": "ball"
    }
  ]
}

You can find some notebooks here.

We provide some Google Colab tutorials:

Notebooks	Google Colab
How to make a simple workflow
How to run Neural Style Transfer
How to train and run YOLO v7 on your datasets
How to use Detectron2 Object Detection

Python API documentation can be found here. You will find Ikomia HUB algorithms code source in our Ikomia HUB GitHub.

This part is coming soon...:)

Distributed under the Apache-2.0 License. See LICENSE.md for more information.

If you use Ikomia in your research, please use the following BibTeX entry.

@misc{DeBa2019Ikomia,
  author =       {Guillaume Demarcq and Ludovic Barusseau},
  title =        {Ikomia},
  howpublished = {\url{https://github.com/Ikomia-dev/IkomiaAPI}},
  year =         {2019}
}

Contributions, issues, and feature requests are welcome! Give a ⭐ if you like this project!

Ikomia - @IkomiaOfficial - [email protected]

Project Link: https://github.com/Ikomia-dev/IkomiaAPI