Assessment of hip joint arthrosis based on the X-ray images of adjacent bones
Scientific paper originated from my engineering thesis, focused on investigating the correlation between structural changes in bones surrounding the hip joint and the joint's degeneration.
Throughout the project, I applied various data normalization methods, image resizing techniques, convolutional neural networks (CNNs), and deep learning approaches to improve the accuracy of arthrosis classification.
Ultimately, I was able to show the correlation between the studied bone regions and hip osteoarthritis, which opens the door to diagnosing the condition without relying on direct imaging of the joint itself.
CycleGan DeepFake
This project aimed to utilize CycleGan architecture and create targeted DeepFake model.
Deepfake is targeted, which means that you need to train it for 2 specific faces.
Project involved creating algorithms for automatic face extraction, preprocessing them, training network and swapping faces in the selected video.
YouTube Video
Limitless_Chat
Environment that allows you to use all the benefits of LLMs without concerns about privacy or answer refusal. It involves GUI, RAG with both online and local search (using vector database).
It allows also for safe search which uses online RAG for improved anonymity. Models are optimized to run on local machine by quantization and half-precision techniques.
Limitless_chat main points:
- No refusal policy
- Safe search
- Local
- Optimization
Currently in development / prototype state.
BobRoss ProGAN
Generative neural network designed to create paintings in the style of Bob Ross.
It is based on the Progressive-GAN architecture, which generates images by gradually increasing their resolution.
The network was implemented from the scratch and was trained on the dataset of 5000 images, painted in the artist's style.
CornHub: Instance-Segmentation dataset of corn cobs
CornHub is a dataset containing corn cobs and their corresponding masks captured in real field conditions. It includes 304 high-resolution RGB images.
It can be used for training and evaluating instance segmentation or object detection models in computer vision, particularly in agriculture as well as in related scientific research.
The dataset is available via one of the following links:
Kaggle
PapersWithCode
NeuroUtils
Programming library focused on automating tasks and managing projects related to computer vision and deep learning.
It was used as base for managing research in the "Assessment of hip joint arthrosis based on the X-ray images of adjacent bones" paper and engineering thesis.
