VOC Segmentation App with Custom U-Net

Pixel-level multi-class segmentation across all 21 Pascal VOC classes.

Discuss a Similar ProjectOpen full report (PDF)
Domain
Computer Vision
Year
2025
Type
Applied build — internal engineering work

This is internal engineering and R&D work, not a client project. It is published to show technical depth — the modelling, evaluation, and tooling behind the AI automation we build for clients. No client is associated with it.

A semantic-segmentation system that classifies every pixel in an image using a custom U-Net (ConvTranspose2d upsampling) on the Pascal VOC 2012 dataset, supporting all 21 classes — wrapped in a visual app for real-time exploration of predictions.

Tools & methods

  • PyTorch Lightning
  • Custom U-Net (Conv2d / ConvTranspose2d)
  • Weights & Biases
  • Matplotlib
Full technical report

Objectives

  • Build a U-Net-style segmentation model from scratch.
  • Support all 21 VOC classes with pixel-wise accuracy.
  • Normalize mask colours and log results live; ship a visual testing tool.

Process

  • Parse and convert RGB mask labels into integer masks; preprocess and normalize.
  • Build encoder–bottleneck–decoder U-Net; train with validation + metric logging.
  • Use ModelCheckpoint to save the top models by validation IoU.
  • Display batches and side-by-side image/mask/prediction grids in the app.

Results

  • Stable training and validation IoU with consistent foreground segmentation (people, dogs, bikes).
  • A 3-panel app comparing input, mask, and prediction per epoch.

What we took from it

  • RGB-mask → class decoding and batch normalization stabilize training.
  • Balancing performance with real-time logging is a core tradeoff.

From the report

  • VOC Segmentation App with Custom U-Net overview
    VOC Segmentation App with Custom U-Net overview

Filter by category

Every project and study. Showing 25 of 26.

Client engagements

  • OpenInterview marketing site shown on a laptop — upload a résumé to generate a shareable video profile
    Client engagement

    US-based HR technology startup · HR technology & recruitment

    OpenInterview

    OpenInterview is a video interview platform that turns a candidate résumé into a recruiter ready profile with a video introduction availability details and one shareable link. Barakode built the product from résumé parsing and video processing to public profiles and Stripe subscriptions.

    Delivered
    4 services
    Timeline
    August 2024 – November 2024
    • React.js
    • Node.js
    • REST API
    • AI résumé processing
    • Cloud storage

    Platform launched and live at openinterview.me, serving job seekers internationally.

    View case study
  • Beyut Libya real-estate platform on a laptop — sign-in screen with the Add New Property wizard behind it
    Client engagement

    Real-estate marketplace operator · Real estate

    Beyut Libya

    Beyut Libya is a bilingual property marketplace for Libya. Buyers can find verified homes with live filters while agents manage their inventory through a secure admin panel. The platform supports residential and commercial property across the country.

    Delivered
    4 services
    Timeline
    October 2024 – January 2025
    • Next.js
    • React.js
    • Node.js
    • REST API
    • MySQL

    Platform live and operating nationwide at beyutlibya.com.

    View case study
  • Gman tailoring platform dashboard shown on screen
    Client engagement

    Growing tailoring & fashion-retail business · Tailoring & fashion retail

    Gman Stitching Platform

    A full-cycle tailoring and order-management platform: customers, measurements, orders, inventory, billing, and accounting in one system. It replaced a PHP/Laravel application that had become too slow to use and too brittle to change.

    Delivered
    3 services
    Timeline
    January 2023 – March 2023
    • React.js
    • Redux
    • Ant Design
    • Node.js
    • Express.js

    Approximately 70% improvement in application response time after the rebuild.

    View case study
  • Barakah Accounts inventory items screen
    Client engagement

    Restaurant operations team · Hospitality & food service

    Barakah Accounts

    A restaurant ERP built to bring orders, stock, billing, reporting, and the day-to-day handover between front-of-house and back-office teams into one working system.

    Delivered
    3 services
    Timeline
    Live product
    • NestJS
    • PostgreSQL
    • Redis

    Live restaurant ERP available at app.barakahaccounts.com.

    View case study
  • Ophir Jewellery purchase orders screen
    Client engagement

    Jewellery business · Jewellery retail & operations

    Ophir Jewellery ERP

    An ERP for a jewellery operation that brings product records, stock, customer activity, order handling, and business reporting into one calmer working view.

    Delivered
    3 services
    Timeline
    Live product
    • NestJS
    • PostgreSQL
    • Redis

    Live jewellery ERP available at ophir.eaquirstech.com.

    View case study
  • Turner 10 sales workspace screen
    Client engagement

    Real-estate operations team · Real estate

    Turner 10 Real Estate ERP

    A real-estate ERP that gives the team one practical place to follow properties, enquiries, availability, client records, and the internal work that supports every deal.

    Delivered
    3 services
    Timeline
    Live product
    • NestJS
    • PostgreSQL
    • Redis

    Live real-estate ERP available at turner10.eaquirstech.com.

    View case study
  • HMS hospital management purchase returns workspace with settlement, credit note, replacement, and refund summaries
    Client engagement

    Hospital operations team · Healthcare administration

    HMS

    HMS brings hospital administration, patient services, billing, pharmacy, and procurement into one operational workspace. Teams can review daily work and follow supplier purchase returns from a clear shared interface.

    Delivered
    3 services
    Timeline
    Hospital project
    • NestJS
    • PostgreSQL
    • Redis
    View case study

Engineering & AI R&D

This is internal and academic engineering work. Each study includes its full technical report.

Need this kind of capability in a real product?

Tell us the workflow you want to automate and we will map the practical version of it — grounded, reviewable, and connected to the tools you already run.

Book a Free Project Discovery CallExplore AI Automation
Contact via WhatsApp+92 300 8698044
Cost Estimator
Message us on WhatsApp