Crate training 

Burn-based training and evaluation for CortenForge detection models.

This crate provides:

• Dataset loading and collation (collate, collate_from_burn_batch).
• Training loop utilities (run_train, TrainArgs).
• Model checkpoint loading/saving helpers.

Supports both LinearClassifier and MultiboxModel from the models crate.

Backend Selection

• backend-wgpu: Uses WGPU for GPU-accelerated training.
• Default: Falls back to NdArray CPU backend.

Stability

Training APIs are experimental and may change as the training pipeline evolves. Core model types (TinyDet, BigDet) are stable, but training utilities and loss functions are subject to refinement.

Re-exports

pub use dataset::collate;

pub use dataset::collate_from_burn_batch;

pub use dataset::CollatedBatch;

pub use dataset::DatasetPathConfig;

pub use dataset::RunSample;

pub use util::run_train;

pub use util::TrainArgs;

Modules

dataset

util

Structs

LinearClassifier

LinearClassifierConfig

MultiboxModel

MultiboxModelConfig

Type Aliases

TrainBackend

Backend alias for training/eval (NdArray by default; WGPU if enabled).