Annotate Images Like a Pro: Python Image Annotation Tool Walkthrough

Image Annotation Made Easy with DigitalSreeni's Python Tool
In this video, I walk you through my Python-based image annotation application and its associated tools, providing a step-by-step demo to help you get started.
Topics Covered:
--Installation of the Python library for image annotation, along with setting up Anaconda and configuring your environment.
- Creating new projects and adding 2D and multi-dimensional images (TIFF, CZI).
- Manual annotation of 2D images and slices from multi-dimensional images using polygon and rectangle tools.
- Semi-automatic annotations with the Segment Anything Model (SAM).
- Renaming and assigning colors to classes for better organization.
- Exporting annotations to various formats: COCO JSON, YOLO v8, labeled images, semantic images, Pascal VOC bounding boxes.
- Verifying exported annotations by reloading them into the program.
Additional Tools:
- Annotation statistics
- Combining JSON annotations
- Data splitting
- Patch extraction
- Data augmentation of images and annotations
Links:
GitHub repository: github.com/bns...
PyPI for pip install info: pypi.org/proje...
To Install:
pip install digitalsreeni-image-annotator
Once installed, simply type sreeni in your command prompt within the correct environment to launch the application.
You can download SAM models from the following links. Please be cautious about the large model on systems with limited memory.
github.com/ult...
github.com/ult...
github.com/ult...
github.com/ult...
It is recommended to place the SAM models in a directory from where you normally start the application to avoid multiple downloads of the same models from the Ultralytics server.

[@TheTimtimtimtam]

Bless your wonderful work Sir, Thank you kindly.

[@DigitalSreeni]

You are very welcome

[@Brickkzz]

This is by far the best tool I’ve used. It’s much easier and more flexible than other online options. The interface is intuitive, and integrates smoothly with my YOLOv8 training workflows. Highly recommend for anyone in computer vision! Few very minor improvements: (1) add automated labelling (e.g. DINO+SAM, or custom YOLO that was pretrained on the dataset), (2) add an option Save As... for the projects, (3) editing the polygon of SAM-labelled instances (for minor corrections).

[@DigitalSreeni]

Thank you very much. 1. Right now we have SAM semi-auto labeling. I intend to add others and thanks for your suggestions. 2. Already added Save As option, need to push the updated code today or tomorrow. 3. Polygons can be edited, just double click on an annotation to get into the edit mode. I have also implemented merge annotation feature which can useful when SAM gets it wrong and you need to manually add part of the object and merge the annotations.

[@ponyspitsfire3035]

Awesome tool! Have you considered adding „auto“ bbox detection using a model such as Grounding DINO?

[@mdFaisal-u3g]

can you add multi-out/ multi label / multi class classification option as well?

[@victorsilvadossantos2769]

Great piece of work! Thanks for sharing this!

[@srivathsansanthanam639]

I DIDN'T FIND A FREEWARE WHICH WAS USER FREINDLY WITH A GOOD UI. I WAS PLANNING TO DESIGN ONE EXACTLY SIMILAR TO THIS AND GIVE IT FREE TO THE COMMUNITY IN A YEAR OR SO. YOU JUST DID IT

[@yogidwitama2480]

Thank you, you always help with my project, sir. I have tried your project, and it's great ,Here’s my input: 1. Add an edit annotation feature. When using SAM-assisted annotation, sometimes the annotated area is not quite accurate, so a function for editing annotations is needed. 2. Add CTRL+Z for undo. 3. Add the ability to hold the mouse wheel to drag the image. 🫡

[@DigitalSreeni]

Thank you. 1. Annotations can be edited, just double click on an annotation to edit. I am also in the process of adding merge annotation feature, so you can manually draw around any mistakes by SAM and merge both annotations. 2. Noted. 3. You can zoom in and pan the image by using mouse, just hold the ctrl button down and use the wheel to zoom in and out or click and move the mouse to pan.

[@yogidwitama2480]

​@@DigitalSreeni Okay, thank you for the explanation on point 1. Is it possible for us to adjust the background opacity?

[@akshatbhatnagar3571]

This is great tool. Thank you

[@DigitalSreeni]

You're welcome!

[@jesussoto8628]

This tool looks great Thanks for sharing

[@DigitalSreeni]

Thanks for watching! I hope you will find it to be useful.

[@coder_zero]

This is sooooooo coooollllllll 😍🤩😍

[@DigitalSreeni]

Thanks

[@Xamy-]

Feedback is put a desaturated white green background and darkened green for the font color. Common trick to enhance readability. Add a confirmation to deleting classes. Allowing a user to choose a yolo model to assist with labelling in the same way as SAM Otherwise interesting stuff :)

[@DigitalSreeni]

Thanks for your feedback. When you were referring to darkened green font color, I assume you men for the slices. I fixed that part and also added confirmation for deleting a class. It used to be there, but somehow lost that functionality. I will release the updated version after a few tests. Using alternate, customized model for annotation assist is on my wish list. I personally like Mask R-CNN (Detectron2).

[@tahirak.7565]

Thankyou.Saved for later. Could you please make Video on yolo 11.

[@raghvendrabhargava8313]

This is a awesome job. I am going to use this right away.

[@DigitalSreeni]

Have fun! :)

[@sanumioluwafemi7247]

Great tool. Can I use this with .bmp images?

[@sanumioluwafemi7247]

Well, I tried it and it worked perfectly

[@DigitalSreeni]

Yes, of course

[@cyberhard]

Very nice! Thanks for the release and the video. The auto segmentation is a great tool. Auto labeling using one's own object detection model would be a great addition. Have you considered supporting ONNX and OpenVINO? They both provide an increase in interference speed over the PyTorch model.

[@DigitalSreeni]

Auto-labeling using own trained model is something on my wish list. In fact, I had it for a couple of versions and had to remove it as it wasn't working well on Linux or mac. Thanks for the suggestion, reaffirms my wishes :) I am relying on Ultralytics for SAM which uses Pytorch, hence the need for it.

[@ngocthienle8828]

Thanks for sharing this tool.

[@DigitalSreeni]

You're welcome!

[@inquisitiverakib5844]

can we get JSON format annotation from it?

[@DigitalSreeni]

Yes.