Data Labeling, Tracking and Annotation with AI

DLTA-AI is the next generation of annotation tools, integrating the power of Computer Vision SOTA models to Labelme in a seamless expirence and intuitive workflow to make creating image datasets easier than ever before

Installation 🛠️

Install Pytorch (for Options 1 and 2)

preferably in a conda environment with python 3.8

install pytorch according to your device from here

conda create -n DLTA-AI python=3.8
conda activate DLTA-AI

<pytorch installation command>
Ex: conda install pytorch torchvision torchaudio .... -c pytorch>

Option 1: Using pip

pip install DLTA-AI

then run it from anywhere using

DLTA-AI

note that first time running DLTA-AI, it will download a required module, it may take some time

Option 2: Manual Installation

Download the lastest release from here

install requirements

pip install -r requirements.txt
mim install mmcv-full==1.7.0

then Run the tool from DLTA_AI_app directory

cd DLTA_AI_app
python __main__.py

Option 3: Using Executable (CPU only)

you can download the lastest release Executable from here it's currently available for windows and linux only, mac version isn't available yet

The Executable doesn't require any installation, just download and run it, however it runs on CPU only (no GPU support) so it's not recommended for large datasets

Solutions to possible problems

click to expand

1. (linux devices 🐧)

some linux machines may have this problem

Could not load the Qt platform plugin "xcb" in "/home/<username>/miniconda3/envs/test/lib/python3.8/site-packages/cv2/qt/plugins" even though it was found.
This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.

Available platform plugins are: xcb, eglfs, linuxfb, minimal, minimalegl, offscreen, vnc, wayland-egl, wayland, wayland-xcomposite-egl, wayland-xcomposite-glx, webgl.

it can be solved simply be installing opencv-headless

pip3 install opencv-python-headless

2. (windows devices 🪟)

some windows machines may have this problem when installing mmdet

Building wheel for pycocotools (setup.py) ... error
...
error: Microsoft Visual C++ 14.0 or greater is required. Get it with "Microsoft C++ Build Tools": https://visualstudio.microsoft.com/visual-cpp-build-tools/

You can try

conda install -c conda-forge pycocotools

or just use Visual Studio installer to Install MSVC v143 - VS 2022 C++ x64/x86 build tools (Latest)**

3. Problem in installing mmcv-full

you may often stuck in installing mmcv-full with this message

Building wheels for collected packages: mmcv-full
  Building wheel for mmcv-full (setup.py) ...

you can try installing pytorch 1.13.1, instead of the lastest version, you can also refer to this isse

Segment Anything 🪄 (NEW)

DLTA-AI takes the Annotation to the next level by integrating lastest Meta models Segment Anything (SAM) to support zero-shot segmentation for any class

SAM can be used also to improve the quality of Segmentation, even inaccurate polygons around the object is enough to be segmented correctly

SAM doesn't only work for Segmentation tasks, it's build in the video mode to support Object Tracking as well for any class

Model Selection 🤖

For model selection, DLTA-AI provides the Model Explorer to utilize the power of the numerous models in mmdetection and ultralytics YOLOv8 as well as the models of SAM

the to give the user the ability to compare, download and select from the library of models

Segmentation 🎨

Using the models from the Model Explorer, DLTA-AI provides a seamless expirence to annotate single image or batch of images, with options to select classes, modify threshold, and full control to edit the segmentation results.

and as mentioned before, **SAM** is fully integrated in DLTA-AI to provide zero-shot segmentation for any class, and to improve the quality of segmentation

Object Tracking 🚗

Built on top of the segmentation and detection models, DLTA-AI provides a complete solution for Object Tracking, with 5 different models for tracking

To impr DLTA-AI have options for video navigation, tracking settings and different visualization options with the ability to export the tracking results to a video file

Beside this, DLTA-AI provides a completely new way to modify the tracking results, including edit and delete propagation across frames

Beside automatic tracking models, DLTA-AI provides different methods of interpolation and filling gaps between frames to fix occlusions and unpredicted behaviors in a semi-automatic way

Export 📤

For Instance Segmentation, DLTA-AI provides to option to export the segmentation to standard COCO format, and the results of tracking to MOT format, and a video file for the tracking results with desired visualization options e.g., show id, bbox, class name, etc.

DLTA-AI provides also the ability to add user-defined or custom export formats that can be used for any purpose, once the user defines his own format, it will be available in the export menu.

Other Features 🌟

Threshold Selection (Confidence and IoU)
Select Classes (from 80 COCO classes) with option to save default classes
Track assigned objects only
Merging models (Run both models and merge the results)
Show Runtime Type (CPU/GPU)
Show GPU Memory Usage
Video Navigation (Frame by Frame, Fast Forward, Fast Backward, Play/Pause)
Light / Dark Theme Support (syncs with OS theme)
Fully Customizable UI (drag and drop, show/hide)
OS Notifications (for long running tasks)
using orjson for faster json serialization
additional script (external) to evaluate the results of segmentation (COCO)
additional script (external) to extract frames from a video file for future use
User shortcuts and preferences settings

Contributing 🤝

DLTA-AI is an open source project and contributions are very welcome, specially in this early stage of development.

you can contribute by:

Create an issue Reporting bugs 🐞 or suggesting new features 🌟 or just give your feedback 📝
Create a pull request to fix bugs or add new features, or just to improve the code quality, optimize performance, documentation, or even just to fix typos
Review pull requests and help with the code review process
Spread the word about DLTA-AI and help us grow the community 🌎, by sharing the project on social media, or just by telling your friends about it

Acknowledgements 🙏

This tool is part of a Graduation Project at Faculty of Engineering, Ain Shams University under the supervision of:

we want also to thank our friends who helped us with testing, feedback and suggestions:

Resources 🌐

License 📜

DLTA-AI is released under the GPLv3 license.

things-wedo/SegmentationEngine