zhenglin 2.1

A deep-learning package contains a set of well-organized deep-neural networks&tools.

pip install zhenglin

@ Author: zhenglin @version: 2.0

This package contains some off-the-shelves deep-learning networks implemented with .

use

pip install zhenglin

to install this package.

Updates

• Apr 18 2025: Version: 2.0. Add init version of supercv(cv) class, a universal wrapper for cv2, PIL.Image and Torch.tensor.
• Nov 13 2024: Version: 1.21. Package overhaul. Add huggingface accelerate, pytorch-lightning, and pytorch-ddp support.
• Oct 26 2023: Version: 1.20.19 Add TransUNet and SwinUNet
• Aug 13 2023: Version: 1.18.15 Add basic quantization support.

Introduction

zhenglin package was motivated by eriklindernoren and his repo which provides many grab-and-run implementation of DL variants. It is the level ground for zhenglin's deep learning projects.

Specifically, this package provides

• universal structure-template:
  • zhenglin.accelerate.train
  • zhenglin.lightning.train
  • zhenglin.pytorch.train
• loss functions and metrics
  • zhenglin.utils.loss
  • zhenglin.utils.metrics
• utils
  • zhenglin.utils.utils
• 20 modular deep-learning networks
  • zhenglin.networks.*

Networks(Alphabetically):

• cycleGAN
• DDPM
• DeblurGAN
• EDSR
• ESRGAN
• Noise2Void
• Pix2Pix
• RCAN
• ResNet
• Restormer
• RRDBNet
• SqueezeNet
• SRDRM
• SRGAN
• SwinIR
• SwinUNet
• TransUNet
• U2Net
• UNet
• Attention-UNet