About Me
My Projects
Deep Learning Infrastructure
Model serving using cheap instances is addressed in this project. The Publish-Subscribe pattern is used to build the infrastructure around Spot instances to server Deep Learning models. Celery is the main tool used to implement the Publish-Subscribe pattern. More details about the project are explained in this blog post.
The Github repository is available here.
Skills used:
- System Design knowledge
- Backend Knowledge (Django, Celery)
- Data Base Knowledge (Race Condition, Lock/Block Queries, Myqsl)
- Server Architecture Knowledge (Linux, Google Cloud, Docker)
- Deep Learning Models Deployment (TensorFlow)
Line Art Colorization Using Deep Learning
This project is a model implementation of the paper called Deep Line Art Video Colorization with a Few References. This deep learning model takes as input line art images and color images where the latter are used as a reference to colorize the former. More details about the project can be found here.
Some of the results from the model.
The Github repository for the model implementation is available here. Additionally, an Electro application built around the model using TensorFlow.JS is available here and the application details can be found here.
Skills used:
- Paper Reading
- Create a Training Workflow
- Understand how to implement models from papers
- Training Data Collection and Creation
- TensorFlow + Javascript (Electron)
Face Detection
The idea behind this project is to create a face detection system suitable for mobile devices. Thus the BlazeFace architecture was used to build the model, this model was trained from scratch using the WIDER FACE: A Face Detection Benchmark and the Face Detection Data Set and Benchmark (FDDB) datasets. The data recollection and creation was a crucial point to obtain good results from the model. The whole journey to train the model can be read in this blog post.
The Github repository for the model implementation is available here.
Two apps for iOS were also created. The first app uses Tensorflow Lite to run the model, and the second app uses CoreML
Skills used:
- Training Data Collection and Creation
- Create a Training Workflow
- Model Deployment
- Swift, iOS applications