Loss function Package Tensorflow Keras PyTOrch

Thanks for the work by: Chen Y, Yu L, Wang J Y, et al. Adaptive Region-Specific Loss for Improved Medical Image Segmentation.

Different Loss Function Implementations in PyTorch and Keras

PyTorch implementation of U-Net and SegNet segmentation models for detecting melanoma, along with pre-trained models and utility functions for evaluation.

Providing a pipeline for training a segmentation model with different architectures

U-Net with ResNet backbone for multi-class segmentation. Includes advanced data augmentation, mixed-precision training (AMP), custom loss functions (Tversky, Dice, BCE), threshold tuning, and evaluation with IoU/Dice metrics. Built in PyTorch with Albumentations, optimized for GPU acceleration and scalable on Vertex AI / GCP.

Сегментация КТ-снимков лёгких для выявления поражения тканей от COVID-19 на основе глубоких свёрточных нейронных сетей