Liveness detection
The projest was developed in collaboration with Identt. The project aims to develop a system that can detect the liveness of the person in front of the camera. Among many methods that can be used to trick the liveness systems 3D masks 2D masks/prints Video replays Deep fakes … we had to prepare a system that will cover them all. My task was to focus on detecting deep fakes. After delivering this part of the system i focused on a relatively new method called rPPG. It tries to measure the pulse of the person by analyzing the color changes in the face.
Oct 26, 2023
Fencing
The projest was developed in collaboration with Nomtek. The project aims to develop a system that can detect the moves that athletes make in real-time. The system will be able to provide feedback on the athletes’ performance and help them improve their skills. The system is able to detect the keypoints of the fencers and track them during the whole fight. Later the collected data is being processed by trained AI system to predict the moves of the fencers. In the end system gathers time series of moves and provides feedback to the fencers. I was coordinating the labeling of the data and developing the machine learning models. Collaborating with Aleksandra Shelton we established a list of most common moves in fencing. Later I gathered all the annotated data, analyzed it and trained the models. In the end I prepared a simple front end to present the results to the fencers.
Oct 26, 2023
Chessgrow - tool to detect chess pieces
In this project, done together with Chessgrow we created a system to detect chess pieces on the image and then convert their coordinates from 2D (image) to 3D (chessboard). The system was able to detect the chess pieces on the image and then convert their coordinates to the 3D chessboard. The goal of the sustem was to connect to the VR headset and allow the user to play chess in VR.
Oct 26, 2023
Audioburst - audio search engine
In this project, done with Audioburst we developed an audio search engine. The system is able to analyze the audio files and provide a search engine that can search for specific words or phrases in the audio files. The system is able to provide the user with a list of audio files that contain the searched words or phrases. The system is also able to provide the user with a list of audio files that are similar to the searched words or phrases. The tasks were very broad - from developing a tool to classify the audio (mostly radio streams and podcasts) to creating a neural net capable of splitting longer audio into smaller parts of different content.
Oct 26, 2023
ABM - tool to diagnose depression
In this project, financed by ABM done together with Stermedia and Affexy are developing an AI system to diagnose depression. The system bases on high quality data input gathered with help of IPIN. The aim of the project is to gather text data (NLP), video data (facial recognition), and audio data (voice analysis) to diagnose depression or at least to be able to suggest a visit to a specialist. In this project I am responsible for developing the Audio part of the system, creating the architecture and data pipelines and collaborating with doctors.
Oct 26, 2023
Abastroke - therapy for stroke patients
In this project, done with Abastroke and Revolve Helthcare we developed an AI system to suggest rehabilitation methods for stroke patients. The app consisted of many different exercises that were tailored to the patient’s needs. The app was able to track the patient’s progress and adjust the exercises accordingly using the AI systems. The app was also able to provide feedback to the patient and the therapist. I was responsible for leading the ML team, creating the architecture and data pipelines, directing the programming efforts of the team, and collaborating with doctors, developers, and UX designers.
Oct 26, 2023