Transformations Meeting 2

Requirements

• 3D - Needed
• 4D - Likely (To be discussed)
• GPU - Needed
• CPU - Needed
• Differentiable - Needed when available
• Parameterisable - Needed
• Invertible - Needed when available
• Consistent UX - Beneficial
• Apply to different pairs - Needed
• Lazy spatial transformations - Beneficial
• Take image geometry into account - Needed
• DAG/Conditionals/Composition - Likely (To be discussed)
• Caching Intermediaries - Likely (To be discussed)
• Reproduciblility for randomised transformations
• API for integration of external packages with working examples

Questions to be discussed next meeting

Dicts vs Lists? DAG vs Compose? Torch Tensor vs numpy?