Crate j2k_ml 

Thin Burn tensor adapter for the j2k owned batch codec.

The codec crates own parsing, grouping, decoding, and accelerator execution. This crate materializes CPU groups or stages completed CUDA and Metal codec output through host memory before an ordinary Burn upload. Casting and normalization stay in ordinary Burn tensor operations after decode.

Re-exports

pub use cpu::CpuBurnDecoder;

pub use cuda::CudaBurnDecoder;

pub use cuda::CudaUploadBurnDecoder;

pub use cuda::SubmittedCudaBurnBatch;

pub use cuda::SubmittedCudaUploadBurnBatch;

pub use metal::MetalBurnDecoder;

pub use metal::MetalUploadBurnDecoder;

pub use metal::SubmittedMetalBurnBatch;

pub use metal::SubmittedMetalUploadBurnBatch;

Modules

cpu

Portable materialization of native codec batch groups into Burn tensors.

cuda

CUDA codec decode followed by an explicit staged Burn upload.

metal

Metal codec decode followed by an explicit staged Burn upload.

Structs

BurnBatchDecode

Successful tensor groups plus indexed preparation and homogeneous execution failures.

BurnBatchGroup

One homogeneous decoded tensor group and its codec metadata.

BurnBatchGroupError

Failure while submitting or completing one homogeneous Burn tensor group.

Enums

BurnBatchTensor

Ordinary rank-4 Burn integer tensor tagged with its exact codec sample type.

BurnDecodeError

Failure while decoding or staging a codec batch into Burn.