From complex data to simple solutions
Personal Projects
Projects and open source contributions I've developed in my spare time
Unpooling Encoder-Decoder
Developed and published a set of 10 deep learning models to the open source kaggle model repository. Received more than 20 upvotes and more than 2000 downloads (click on the title to access the models).
The models represent image encoders and decoders trained on the COCO 2017 dataset where encoders are VGG19 feature extractors from various intermediary layers that store argmax information from max-pooling operations and decoders mirror the architecture of the encoders and use unpooling layers for upsampling.
Image Style Transfer Python Library
[Link to gitlab repository]
[Link to pypi]
An AI toolkit that transfers artistic styles between images using neural networks.
The project implements three different methods named after the first author of their corresponding original paper (Gatys, Yijun, Reinhard) to apply the visual style of one image to the content of another image. It runs relatively fast (under 30 seconds with a GPU) and includes a Python library called "stram" (style transfer machine) for easy integration into other applications.
Essentially, it lets you take the artistic style from one picture and apply it to a completely different picture while preserving the original content.
Image Style Transfer Web Application
[Link to gitlab repository for frontend]
[Link to gitlab repository for backend]
A pair of code repos that work together to deploy a web application (only locally for now) that one can use to pre-process and apply style transfer to custom images.
It optimizes the data processing steps by caching previous image states and uses the package described in the previous section.
The tech stack used includes Django and React.
A small demo is shown below ⬇️.
Image Style Transfer Experiments
[Link to gitlab repository]
[Link to pypi]
An AI toolkit that transfers artistic styles between images using neural networks.
The project implements three different methods named after the first author of their corresponding original paper (Gatys, Yijun, Reinhard) to apply the visual style of one image to the content of another image. It runs relatively fast (under 30 seconds with a GPU) and includes a Python library called "stram" (style transfer machine) for easy integration into other applications.
Essentially, it lets you take the artistic style from one picture and apply it to a completely different picture while preserving the original content.
