areyoustillreading
Things I built

Work that earned its writing.

Selected engineering and research work — the kind of project I couldn't write about honestly until I'd built it first. One thing done properly, rather than a grid of repositories nobody opens.

MaintainedOpen source

Deep-Learning Architectures from Scratch

An open-source PyTorch library implementing 25+ computer-vision architectures — CNNs, vision transformers and segmentation networks — reimplemented from their original papers.

pytorchcomputer-visionopen-sourcecv 0.0.1
GitHubDocumentation →

Architecture families

25+architectures, each traced back to its paper
  • CNNsResNet, ConvNeXt, DenseNet, AlexNet, Inception-v1
  • TransformersDeiT, Swin, Deep ViT, Bottleneck ViT, CeiT, ConViT
  • SegmentationSegmentation networks

How it's built

  • Modular, packaged codebase
  • Unit tests
  • GitHub Actions CI/CD
  • Config-driven training
  • Experiment logging
  • Hosted documentation
Open source

Memoir

A project to colorize images

pytorchgans
GitHub

This is where the essays come fromThe reimplementations are what make the writing possible — you can't explain an architecture honestly until you've had to make it train.

Read the essays →
Everything here is open source — MIT unless the repository says otherwise.More on GitHub · github.com/shivam7569 →