Skip to main content

LIS Sign Language Recognition

·
Academic Personal Python PyTorch Computer Vision CNN OpenCV Torchvision TensorBoard
University Projects - This article is part of a series.
Part 1: This Article
A computer-vision project for recognizing the LIS sign-language alphabet using convolutional neural networks and a custom data workflow.
Type Academic machine-learning project

Built around model design, training, and the practical side of dataset preparation.

Focus Recognition pipeline

Model architecture, data collection, preprocessing, training, and evaluation.

Stack Python and PyTorch

Deep-learning experimentation with supporting OpenCV and TensorBoard tooling.

GitHub
Python
PyTorch
OpenCV
Torchvision
TensorBoard

Problem Framing
#

The project goal was to recognize signs from the LIS alphabet through a computer-vision pipeline. That makes it a model problem, but not only a model problem. Accuracy depends heavily on how data is collected, organized, and cleaned before training ever begins.

That is why this project matters to me more than a simple “trained a model” summary. The interesting work was in building the surrounding pipeline well enough that the experiments meant something.

Model Work
#

I implemented both custom CNN architectures and AlexNet-based variants, comparing approaches rather than treating the first solution as final.

The practical goal was to understand the tradeoff space:

  • how much complexity the task actually needed
  • how the models behaved under different data assumptions
  • what architecture decisions improved generalization instead of only fitting the training set

Dataset and Tooling
#

Data acquisition and data engineering were a core part of the project. In machine-learning work, that stage often determines whether the modeling phase is credible or just decorative.

I built the pipeline around:

  • dataset preparation for training and evaluation
  • image handling and preprocessing support
  • experiment visibility through tooling such as TensorBoard

That combination made it possible to iterate with feedback rather than treat training as a black box.

What This Project Shows
#

This article belongs in the portfolio because it demonstrates comfort outside game-engine work. It shows that I can move into a different technical domain, reason about pipelines end to end, and structure experiments in a way that is useful rather than superficial.

University Projects - This article is part of a series.
Part 1: This Article