Skip to content
Computer Vision Workbench — Hakimi Sa’ari

Computer Vision Workbench

Manage datasets, inspect data quality, train models, and review predictions in one workspace.

Machine learning tooling · Personal project by Hakimi Sa’ari

Computer Vision Workbench showing a successful demo evaluation, green model-ready result, example mAP score, feedback, and Open Playground action beneath a visible demo-data notice.
Successful Training Lab evaluation with configuration feedback and a next step into Model Playground. Illustrative demo metrics and feedback, not a benchmark.

Open screenshot in the desktop Image Viewer

About this project

An integrated computer vision platform organized around a Data Studio, Training Lab, and Model Playground. It supports detection and classification datasets, data-quality auditing, guided YOLO training, and visual inspection of trained-model predictions.

What I built

  • Built dataset uploads, classification folders, annotation management, and label-repair workflows.
  • Integrated dataset audits using Cleanlab and feature-based cross-validation.
  • Created a guided training flow with model registration and AI-assisted diagnostics.
  • Built an inference playground with visual overlays and raw prediction inspection.

Tools

Python · Flask · SQLAlchemy · PyTorch · Ultralytics · Cleanlab · scikit-learn · Gemini

Source repository is private.

All personal projects · Open on the desktop