Back to All Projects
Case Study • [03.03]
03Computer Vision & Deep LearningInteractive Engineering Demo

CardecaAlgo

RT-DETR & EfficientNet Vehicle Intelligence

RT-DETR Detection TransformerPyTorch EfficientNetInteractive Konva CanvasExportable Inspection Reports
SYSTEM ARCHITECTURE & VERIFIED WORKFLOWLIVE INTERACTIVE DEPLOYMENT
CardecaAlgo • RT-DETR Vision Transformer & PyTorch Inference
60 FPS INFERENCE • 16.4ms
SUV • 98.7%
[124, 76, 492, 340] • IOU: 0.94
SEDAN • 96.2%
[510, 110, 720, 320] • EfficientNet
STREAM: 1080p60RESNET-50 BACKBONE
HTML5 KONVA CANVAS
LATENCY
16.4ms / frame
DETECTION HEAD
RT-DETR Transformer
CLASSIFIER
PyTorch EfficientNet
EXPORT FORMAT
ONNX & TensorRT
01 • Operational Challenge

The Problem & Requirements

Standard vehicle recognition systems either sacrifice inference speed for accuracy or lack fine-grained classification capabilities when inspecting varied vehicle makes, models, and orientations.

02 • Architectural Solution

System Design & Delivery

Implemented a two-tier deep learning inference architecture. An RT-DETR model detects and localizes vehicles in real-time, while a tailored PyTorch EfficientNet classifier determines detailed vehicle attributes, rendered interactively on an HTML5 Konva canvas.

03 • System Architecture & Implementation Details

Technical Pipeline & Data Flow

Next.js App Router frontend; PyTorch inference service deploying RT-DETR and EfficientNet weights; interactive Konva canvas for client-side bounding box visualization and annotation; client-side jsPDF report generation.

Verified Production Features:
Real-time object localization using RT-DETR detection transformers.
Fine-grained model and attribute classification via PyTorch EfficientNet.
Interactive Konva canvas enabling zooming, panning, and bounding-box inspection.
One-click inspection report export in PDF format via jsPDF.
Modern dark-mode interface with Framer Motion visual state transitions.
04 • Verified Technologies & Tooling
Next.js App RouterTypeScriptPyTorchRT-DETREfficientNet-B4KonvajsPDFFramer MotionTailwind CSS v4
LET'S CONNECT • PROJECT INQUIRY

Interested in deploying this architecture or collaborating?

Available for software engineering roles, full-stack systems design, and applied AI application development.

Connect via Direct Channels