traiNNer-redux documentation

Guides

• Getting Started
  • Initial Setup
  • Training a Model
  • Convert models to ONNX
• Training Guidelines
  • Training Schedule

Reference

• Config file reference
  • Top level options
  • Dataset options (datasets.train and datasets.val)
  • Path options (path)
  • Train options (train)
  • Scheduler options (train.scheduler)
  • Validation options (val)
  • Logging options (logger)
• Architecture reference
  • Generator architectures (network_g)
  • Discriminator architectures (network_d)
• Loss reference
  • adistsloss
  • averageloss
  • bicubicloss
  • charbonnierloss
  • colorloss
  • contextualloss
  • distsloss
  • ffloss
  • ganloss
  • hsluvloss
  • l1loss
  • ldlloss
  • lumaloss
  • mseloss
  • mssimloss
  • msssiml1loss
  • multiscaleganloss
  • nccloss
  • perceptualfp16loss
  • perceptualloss

Benchmarks

• PyTorch Inference Benchmarks by Architecture (AMP & channels last)
  • By Scale
  • By Architecture
• PyTorch Inference Speed vs VRAM vs PSNR Charts
  • 4x
  • 3x
  • 2x

Developers

• Contributing
  • Prerequisites
  • Launch training code
  • PR Requirements