Computer Vision Research & Engineering

Real-World Vision
Systems That Deploy.

Lightweight deep learning for semantic segmentation, object detection, and edge AI across intelligent transportation, UAV analytics, medical imaging, industrial inspection, and aviation safety.

Start a Project Discussion View Selected Projects
11+
Years Research Experience
29+
Research Contributions
15+
Vision Models Developed
Multi
Application Domains
Inference Demos

Model Inference in Action.

Selected Projects

Published Architectures.

Research-driven lightweight architectures designed for deployment-aware computer vision systems where accuracy, efficiency, and real-world constraints matter.

View All Selected Projects
Applications

Vision AI Across
Real-World Domains.

Some of the application domains where deployment-aware computer vision systems can support real-world analysis, monitoring, and automation.

Infrastructure Monitoring
Infrastructure AI

Infrastructure Monitoring

Deployment-aware segmentation and detection systems can support inspection workflows by localizing surface defects, road...

Learn More →
Medical Imaging
Medical AI

Medical Imaging

Lightweight segmentation and detection systems can support boundary-aware analysis in clinical imagery, helping demonstr...

Learn More →
UAV & Aerial Analytics
Aerial Vision

UAV & Aerial Analytics

Aerial imagery requires efficient models that can handle scale variation, complex backgrounds, and deployment constraint...

Learn More →
Intelligent Transportation
Transportation AI

Intelligent Transportation

Deployment-aware visual perception supports traffic analysis, roadway monitoring, and safety-critical transportation sce...

Learn More →
AI for Environmental Monitoring
Environmental AI

AI for Environmental Monitoring

Environmental monitoring benefits from lightweight models that can identify visual risk patterns across open-world scene...

Learn More →
Infrastructure Monitoring
Infrastructure AI

Infrastructure Monitoring

Deployment-aware segmentation and detection systems can support inspection workflows by localizing surface defects, road...

Learn More →
Medical Imaging
Medical AI

Medical Imaging

Lightweight segmentation and detection systems can support boundary-aware analysis in clinical imagery, helping demonstr...

Learn More →
UAV & Aerial Analytics
Aerial Vision

UAV & Aerial Analytics

Aerial imagery requires efficient models that can handle scale variation, complex backgrounds, and deployment constraint...

Learn More →
Intelligent Transportation
Transportation AI

Intelligent Transportation

Deployment-aware visual perception supports traffic analysis, roadway monitoring, and safety-critical transportation sce...

Learn More →
AI for Environmental Monitoring
Environmental AI

AI for Environmental Monitoring

Environmental monitoring benefits from lightweight models that can identify visual risk patterns across open-world scene...

Learn More →

View Selected Projects →
Computer Vision Research & Engineering

Lightweight Vision Systems
From Architecture Design to Inference.

Semantic Segmentation

Lightweight encoder-decoder systems for pixel-level analysis across medical imaging, infrastructure monitoring, aerial analytics, and vision-driven applications.

LiteFusionNet DFF-UNet

Object Detection

Efficient detection systems for small objects, visual monitoring, and constrained computer vision applications.

VDXNet LiteFODNet

Lightweight Architecture

Parameter-efficient architectures balancing accuracy, latency, computational cost, and memory constraints for efficient vision systems.

Parameter Efficiency Low GFLOPs

Inference Optimization

TensorRT optimization, ONNX deployment workflows, FP32/FP16 benchmarking, and hardware-aware performance evaluation for embedded vision systems.

TensorRT Jetson
Blog

Technical Articles
Insights, Engineering, & Experiments.

Debugging

Why Your Segmentation Model Predicts Only Background

Root cause analysis of foreground collapse under cross-entropy on imbalanced datasets. CVC-ClinicDB experiment over 100 epochs.

 Practical Guide

How to Train an Object Detection Model on a Custom Dataset

The decisions tutorials skip, dataset format, model scale, evaluation methodology. YOLOv12 on VisDrone.

 Architecture

U-Net Explained Clearly With a Practical Training Example

Encoder, skip connections, and decoder from first principles. CVC-ClinicDB training results across 100 epochs.

 Debugging

Common Mistakes in Object Detection Training That Kill Performance

Wrong annotations, bad augmentation, poor evaluation, and overfitting, each with symptom patterns and fixes.
Read All Articles
Books

Beginner to Advanced.
Two Tracks.

Semantic Segmentation Track
A Step-by-Step Guide for BeginnersBeginner
Training Models on Real-World DataIntermediate
Advanced Architecture Design and ExperimentsAdvanced
Object Detection Track
A Step-by-Step Guide for BeginnersBeginner
Training Models on Real-World DatasetsIntermediate
Advanced Architecture Design and ExperimentsAdvanced
Browse Books
News & Announcements

Latest News.

Tool Release

Dataset Inspector is now available for free.

Explore Dataset Inspector →

 Announcement

OcculinsLab becomes the dedicated home for our computer vision work.

About OcculinsLab →
View all news →

Discuss a Computer Vision Project

Whether you are improving a lightweight architecture, optimizing deployment performance, designing a segmentation system, or exploring a computer vision solution. Occulins provides focused technical consultation and development support.

Schedule Consultation Start a Discussion