Skip to content

Welcome to ATSVisionOps

Python 3.12 YOLO Support PaddlePaddle Support GitCode Release 123 Cloud Drive Support Vercel Support QQ Contact

ATSVisionOps is an integrated data preparation tool for industrial vision projects, covering the complete workflow of annotation → augmentation → format conversion → training → offline distribution.

Quick Start


Why Choose ATSVisionOps

In an industrial vision project, the most time-consuming part is not model tuning, but data preparation:

  • Annotating 10,000 samples takes weeks, the annotation tools are scattered across multiple windows, and the state is inconsistent
  • Dataset formats are converted back and forth between YOLO / COCO / LabelMe, and every toolchain has to be re-run
  • The samples collected on site are unbalanced and need augmentation, but the augmentation software does not talk to the annotation tool
  • A trained model has to be delivered, but the site has no Python environment and requires a pile of dependencies to be installed

ATSVisionOps brings all these tasks that you "have to do but find annoying" into a single window.


Core Capabilities

  • Three annotation types: rectangle / polygon / rotated box
  • Shortcut keys + magnifier + multi-level undo
  • Batch renaming and auto classification
  • Supports multiple formats: JSON / TXT
  • Geometric transformations: rotation, scaling, flipping, shear
  • Color jitter: brightness, contrast, hue, saturation
  • Noise and occlusion: simulate real acquisition environments
  • Labels are synchronized automatically after augmentation
  • Two-way conversion between LabelMe JSON ↔ YOLO TXT
  • LabelMe JSON ↔ COCO JSON
  • Intelligent polygon merging (solves fragmented label problems)
  • Supports three shapes: rectangle / polygon / OBB
  • Circular image flatten and crop (top-view samples made of metal, glass, plastic, etc.)
  • Batch resizing + format conversion
  • Filename sanitizer tool
  • Built-in training panel
  • Real-time TensorBoard visualization
  • One-click environment deployment (online / offline package)
  • One-click installation of offline .whl packages
  • Suitable for production sites with no network access

First launch

  1. Choose your interface language and either the Colorful or Minimal theme, then click Get started.
  2. Module overview opens the first time you reach the main window. Read what each tool does and click its name to open it. To return, use Module overview at the top right of the home page.
  3. To change either setting later, open the gear menu at the top of the sidebar and choose Language or Interface theme. Changes take effect after restarting the app.

Getting Started by Goal

  • : Want to annotate?


    Draw boxes / polygons / rotated boxes on your images from scratch.

    Image Annotation

  • : Need to clean up your data?


    Organize, filter, classify, and batch-process raw datasets.

    Data Optimization

  • : Need data augmentation to expand your samples?


    Geometric / color / noise transformations, with labels synchronized automatically.

    Data Augmentation

  • : How do I generate COCO?


    Convert between the three formats YOLO / LabelMe / COCO.

    COCO Conversion


Community and Feedback

Author's Note

"The original intention behind building these tools was to hand off those fragmented, repetitive data-preparation tasks on the industrial floor — tasks nobody wants to do but that cannot be avoided — to the machine. So that people can focus on what truly matters — the data and the algorithms."

For more on the project's original intention, development journey, and acknowledgements, please read About ATSVisionOps.


Distribution

ATSVisionOps now uses a unified Standard Edition + optional inference components distribution. Basic annotation needs no inference component; install the appropriate component when you need model auto annotation.

The Pro Edition is officially archived and will no longer receive updates as a separate product. here.