Home Page - Sergej Chicherin
Welcome to my homepage, where I collect links to my projects and share various texts.
Computer Works
Here is a curated list of not-so-big projects where the main idea, technical realization, or both were original and mine. The technical part was typically completed in a couple of days or less than a month. Nevertheless, the concepts behind these projects are endlessly interesting for discussion. Some of the projects were done as contract jobs, some were collaborative efforts. Many thanks to everyone who played a part in making these projects happen.
-
2023
Virtual Photo Booth
What: An iOS application that guides users in capturing, removing the background, and cropping portraits for document photos.
Update: In progress.
Computer VisionMobile
-
2023
Cached Dataloader
Concept: This simple tool create distributed in-memory cache, If distributed filesystem is too slow and memory and CPU is good enough,it could significantly speedup training neural networks.
What: It could store unpacked jpeg images as well and probe system memory for availability
AITools
-
2021
Hide Face
What: Anonymizes faces in videos. I optimized a neural network and developed an algorithm for an Android application.
Computer VisionMobile
-
2017, 2023
Emotional Fingerprint from Messages
Concept: An approach to the Turing test applied to individuals. The AI, which is a text simulation of your own personality.
What: Thanks to GDPR, all your personal messages belong to you. You can download private messages from platforms like Facebook and train a neural network (RNN) to mimic your writing style and some thoughts based on incoming messages.
Update(2023): Planning to fine-tune the Large Language Model for this task.
Natural Language Processing
-
2021
Neural Architecture Search on Target Device
What: Predicting efficient neural architectures for Vivante and Hexagon NPU, considering optimization techniques including reparametrization.
Graph Neural NetworksMobile
-
2019
This Fox Doesn't Exist
Concept: One of spin off of this person does not exist targeting artificial foxes drawing
What:I collected and cropped the dataset of foxes (before AFHQ), trained StyleGAN on them and published samples as a web page showing random fox
Computer Vision Demo
-
2019
Neural Avatar Demo
What: Real-time puppetry over a person's face using information from your camery.
Where: Part of my work at Samsung AI was creating demos to showcase how new technologies may look in real life. I developed a web application that uses generative networks for inference and another demo on mobile to show it could to the CEO of Samsung Electronics.
Update: I recommend trying the mature Avatarify project developed by my former colleagues.
Mobile Computer Vision Demo
-
2018
Breast Cancer Detection
What: Detection of breast cancer using an ensemble of neural networks.
Where: Our collective results were published in the Journal of Medical Imaging, Vol. 8, Issue 3. The average precision aligned with that of human pathologists.
Computer Vision
-
2017
Menu Recommendations Based on User Preferences
Concept: Alchemical preferences in food based on multiple hypotheses.
What: Parsing menu dishes into ingredients and classifying into cuisines, style of cooking, chemicals.
AI Demo
-
2017
Video Super Resolution
What: Restoration of old videos based on noise profile analysis.
Where: Presented at Huawei's demo stand at computer vision conference.
Computer Vision Demo
-
2016
Human Face as a Map
Concept: Stylization of a human face resembling a geographical map.
What: Unsupervised implementation of CycleGAN for Face <-> Satellite Images domain transfer.
Where: Showcased at a science art festival, including a lecture with a 500 people audience.
Art Installation Computer Vision
-
2016
Andy vs IRS
What: Optimizing taxes bot that searches for existing optimization strategies based on user input and the US state.
Where: Presented at YC Demo.
Natural Language Processing Demo
-
2015
Glass Bead Fields
Concept: Transferring ideas from a game board into music.
What: The camera captures the game and people, detects and understands positions on the board, and triggers MIDI music events in a variative music composition by Pixelord.
Where: The installation was showcased at the Afisha Festival, the Jewish Museum, and the Parallel Program of the Moscow Biennale in collaboration with MediaMead Studio.
Art Installation Computer Vision
-
2015
Science Article Taxonomy
Concept: Building a scientometric tool for taxonomy and aggregating articles.
What: Unsupervised clustering of papers based on modified LDA.
Where: Internal tool for electronic library.
Machine Learning Natural Language Processing
-
2012
Web interface to Go
Concept:HTML5 Go(igo,baduk,weiqi)
What: Inline Go rules in JavaScript to play online go
AI Demo
-
2011
Face Recognition Module
Concept: Building a web face recognition module.
What: Using OpenCV Eigenfaces, build a module for Java Enterprise Authentification which could be plug-in in any Java web app.
Where: In IBM internal hackathon.
AI Demo
Contact me
Feel free to get in touch:
serenkiy@ieee.org
@chroneus
Sergei Chicherin
@chroneus