j2k-ml 0.9.0

Thin Burn tensor adapter for JPEG 2000 and HTJ2K batch decoding
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138

// SPDX-License-Identifier: MIT OR Apache-2.0

//! Host staging followed by ordinary Burn tensor materialization.

use burn_core::tensor::{backend::Backend, DType, Tensor, TensorData};
use j2k::{BatchGroupInfo, J2kDecodeWarning, NativeSampleType, Rect};

use crate::batch_contract::tensor_shape;
use crate::{BurnBatchGroup, BurnBatchTensor, BurnDecodeError};

pub(crate) fn byte_len(batch: usize, info: &BatchGroupInfo) -> Result<usize, BurnDecodeError> {
    let samples = tensor_shape(batch, info)?
        .into_iter()
        .try_fold(1usize, usize::checked_mul)
        .ok_or(BurnDecodeError::SizeOverflow)?;
    let bytes_per_sample = match info.sample_type {
        NativeSampleType::U8 => 1,
        NativeSampleType::U16 | NativeSampleType::I16 => 2,
        _ => return Err(BurnDecodeError::UnsupportedCodecContract),
    };
    samples
        .checked_mul(bytes_per_sample)
        .ok_or(BurnDecodeError::SizeOverflow)
}

pub(crate) fn materialize<B: Backend>(
    info: BatchGroupInfo,
    source_indices: Vec<usize>,
    decoded_rects: Vec<Rect>,
    warnings: Vec<Vec<J2kDecodeWarning>>,
    bytes: Vec<u8>,
    device: &B::Device,
) -> Result<BurnBatchGroup<B>, BurnDecodeError> {
    let shape = tensor_shape(source_indices.len(), &info)?;
    let expected = byte_len(source_indices.len(), &info)?;
    if bytes.len() != expected {
        return Err(BurnDecodeError::AcceleratorInterop {
            backend: "codec staging",
            message: format!(
                "staged codec output has {} bytes; expected {expected}",
                bytes.len()
            ),
        });
    }
    let tensor = match info.sample_type {
        NativeSampleType::U8 => BurnBatchTensor::U8(Tensor::from_data(
            TensorData::new(bytes, shape),
            (device, DType::U8),
        )),
        NativeSampleType::U16 => {
            let samples = bytes
                .chunks_exact(2)
                .map(|bytes| u16::from_ne_bytes([bytes[0], bytes[1]]))
                .collect();
            BurnBatchTensor::U16(Tensor::from_data(
                TensorData::new(samples, shape),
                (device, DType::U16),
            ))
        }
        NativeSampleType::I16 => {
            let samples = bytes
                .chunks_exact(2)
                .map(|bytes| i16::from_ne_bytes([bytes[0], bytes[1]]))
                .collect();
            BurnBatchTensor::I16(Tensor::from_data(
                TensorData::new(samples, shape),
                (device, DType::I16),
            ))
        }
        _ => return Err(BurnDecodeError::UnsupportedCodecContract),
    };
    Ok(BurnBatchGroup {
        tensor,
        info,
        source_indices,
        decoded_rects,
        warnings,
    })
}

#[cfg(test)]
mod tests {
    #[cfg(not(all(target_arch = "aarch64", target_os = "linux")))]
    use burn_flex::{Flex, FlexDevice};
    #[cfg(all(target_arch = "aarch64", target_os = "linux"))]
    use burn_ndarray::NdArrayDevice::Cpu as FlexDevice;
    use j2k::{
        BatchAlpha, BatchCodecRoute, BatchColor, BatchLayout, BatchWaveletTransform,
        CompressedPayloadKind, CompressedTransferSyntax,
    };
    use j2k_core::Colorspace;

    use super::{byte_len, materialize};
    use crate::BurnDecodeError;

    #[cfg(all(target_arch = "aarch64", target_os = "linux"))]
    type Flex = burn_ndarray::NdArray<f32, i64, i8>;

    fn test_info() -> j2k::BatchGroupInfo {
        j2k::BatchGroupInfo {
            dimensions: (3, 2),
            sample_type: j2k::NativeSampleType::U16,
            precision: 12,
            signed: false,
            color: BatchColor::Rgb,
            alpha: BatchAlpha::None,
            transform: BatchWaveletTransform::Reversible53,
            route: BatchCodecRoute::Htj2k,
            layout: BatchLayout::Nchw,
            colorspace: Colorspace::SRgb,
            transfer_syntax: CompressedTransferSyntax::HtJpeg2000Lossless,
            payload_kind: CompressedPayloadKind::Jpeg2000Codestream,
        }
    }

    #[test]
    fn staged_byte_length_tracks_native_width_and_layout() {
        assert_eq!(byte_len(4, &test_info()).unwrap(), 4 * 3 * 2 * 3 * 2);
    }

    #[test]
    fn staged_size_mismatch_uses_the_existing_interop_error_contract() {
        let result = materialize::<Flex>(
            test_info(),
            vec![0],
            Vec::new(),
            vec![Vec::new()],
            vec![0; 35],
            &FlexDevice,
        );

        let Err(BurnDecodeError::AcceleratorInterop { backend, message }) = result else {
            panic!("size mismatch must use the patch-compatible interop variant");
        };
        assert_eq!(backend, "codec staging");
        assert_eq!(message, "staged codec output has 35 bytes; expected 36");
    }
}