Embedded Deep Learning Semantic Segmentation

TensorFlow Lite Computer Vision Embedded Systems

Created a vision-based environment surveying application using deep learning, specifically optimized to run efficiently on resource-constrained embedded devices.

System Workflow

Workflow Diagram

Training Specifications

Task

Semantic Segmentation

Dataset

CamVid

Architecture

Encoder-Decoder
(UNet, SegNet)

Encoder

MobileNetV2

Optimization Strategy

Technique Post-Training Quantization
Framework TensorFlow Lite
Optimization Process

Compression Results

Model Size Comparison

Achieved significant reduction in model size with minimal loss in accuracy.

Live Inference using TFLite Interpreter

Real-time Inference
Real-time semantic segmentation running on embedded hardware
Back to Projects