git projects

the algorithms and datasets project pages

algorithms
project thumbnail

Policy Gradient-Driven Noise Mask

A novel deep learning approach that employs reinforcement learning to generate adaptive noise masks for medical image classification. This innovative pretraining pipeline combines a lightweight policy network with a classifier network to tackle the challenges of heterogeneous multi-modal and multi-organ biomedical datasets.
project thumbnail

Variational Contrastive Learning

A novel self-supervised learning framework that combines variational methods with contrastive learning to tackle complex classification pretraining tasks. This innovative approach leverages beta-divergence and variational principles to create robust semantic representations, particularly effective for face understanding tasks.

datasets
project thumbnail

Refined RadImageNet

A sophisticated toolkit for processing and enhancing the RadImageNet medical imaging dataset, featuring advanced data organization and model benchmarking capabilities. This conversion tool transforms raw radiological images into a refined and stratified structure, optimized for deep learning applications.
project thumbnail

YFCC392K Face Attributes Dataset

A comprehensive face attributes dataset containing 392,000 diverse facial images curated from YFCC100M, specifically designed for advancing self-supervised and weakly-supervised face recognition research. This carefully processed collection features weakly-labeled facial attributes, enabling robust model training and evaluation.