barkin

PSPNet-Replication is a faithful implementation of the Pyramid Scene Parsing Network designed for semantic segmentation. It reproduces the original paper’s architecture, including a dilated ResNet backbone and an inline pyramid pooling module for multi-scale global context aggregation.

DoubleU-Net is a two-stage U-Net architecture designed for medical image segmentation. It refines the initial prediction by multiplying the input with the first mask and using a second U-Net to produce a more accurate final mask.

ResUNetPlusPlus-Replication is a PyTorch-based implementation of the ResUNet++ architecture designed for medical image segmentation. It focuses on reproducing the model’s core components—residual connections, squeeze-and-excitation blocks, ASPP, and attention mechanisms—in a clean and modular structure.

UNet3Plus-Replication implements the UNet 3+: A Full-Scale Connected U-Net for Medical Image Segmentation architecture in a modular way, focusing on full-scale feature aggregation. It provides a practical translation of the model without training or evaluation components.

Attention-UNet-Replication implements the Attention U-Net architecture for medical image segmentation, integrating additive soft-attention gates into skip connections to highlight relevant features. It is designed to replicate the original model’s structure, mathematics, and block diagram for research and experimentation purposes.

Pix2Pix-Replication implements a conditional GAN framework to translate input images into corresponding outputs using a U-Net generator and a PatchGAN discriminator. It follows the original formulation by combining adversarial loss with L1 loss to produce realistic and structurally accurate images.