TCSegNet:An Anti-Biased TBSRTC-Category Aware Nuclei Segmentation Framework with A Multi-Label Thyroid Cytology Benchmark
This project consists of models of our newly-proposed TCSegNet and its semi-supervised extension, which :
-
Help reduce bias in the learning process of the segmentation model with the routine unbalanced training set.
-
Leverage image-wise labels in a semi-supervised learning manner, which significantly reduces the reliance on annotation-intensive pixel-wise labels.
An dataset of thyroid cytopathology image patches will be released upon acceptance.