tch 0.24.0

Rust wrappers for the PyTorch C++ api (libtorch).
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
139
140
141
142
143
144
145
146

//! Utility functions to manipulate images.
use crate::wrappers::image::{load_hwc, load_hwc_from_mem, resize_hwc, save_hwc};
use crate::{Device, TchError, Tensor};
use std::io;
use std::path::Path;

pub(crate) fn hwc_to_chw(tensor: &Tensor) -> Tensor {
    tensor.permute([2, 0, 1])
}

pub(crate) fn chw_to_hwc(tensor: &Tensor) -> Tensor {
    tensor.permute([1, 2, 0])
}

/// Loads an image from a file.
///
/// On success returns a tensor of shape [channel, height, width].
pub fn load<T: AsRef<Path>>(path: T) -> Result<Tensor, TchError> {
    let tensor = load_hwc(path)?;
    Ok(hwc_to_chw(&tensor))
}

/// Loads an image from memory.
///
/// On success returns a tensor of shape [channel, height, width].
pub fn load_from_memory(img_data: &[u8]) -> Result<Tensor, TchError> {
    let tensor = load_hwc_from_mem(img_data)?;
    Ok(hwc_to_chw(&tensor))
}

/// Saves an image to a file.
///
/// This expects as input a tensor of shape [channel, height, width].
/// The image format is based on the filename suffix, supported suffixes
/// are jpg, png, tga, and bmp.
/// The tensor input should be of kind UInt8 with values ranging from
/// 0 to 255.
pub fn save<T: AsRef<Path>>(t: &Tensor, path: T) -> Result<(), TchError> {
    let t = t.to_kind(crate::Kind::Uint8);
    match t.size().as_slice() {
        [1, _, _, _] => save_hwc(&chw_to_hwc(&t.squeeze_dim(0)).to_device(Device::Cpu), path),
        [_, _, _] => save_hwc(&chw_to_hwc(&t).to_device(Device::Cpu), path),
        sz => Err(TchError::FileFormat(format!("unexpected size for image tensor {sz:?}"))),
    }
}

/// Resizes an image.
///
/// This expects as input a tensor of shape [channel, height, width] and returns
/// a tensor of shape [channel, out_h, out_w].
pub fn resize(t: &Tensor, out_w: i64, out_h: i64) -> Result<Tensor, TchError> {
    Ok(hwc_to_chw(&resize_hwc(&chw_to_hwc(t), out_w, out_h)?))
}

pub fn resize_preserve_aspect_ratio_hwc(
    t: &Tensor,
    out_w: i64,
    out_h: i64,
) -> Result<Tensor, TchError> {
    let tensor_size = t.size();
    let (w, h) = (tensor_size[0], tensor_size[1]);
    if w * out_h == h * out_w {
        Ok(hwc_to_chw(&resize_hwc(t, out_w, out_h)?))
    } else {
        let (resize_w, resize_h) = {
            let ratio_w = out_w as f64 / w as f64;
            let ratio_h = out_h as f64 / h as f64;
            let ratio = ratio_w.max(ratio_h);
            ((ratio * h as f64) as i64, (ratio * w as f64) as i64)
        };
        let resize_w = i64::max(resize_w, out_w);
        let resize_h = i64::max(resize_h, out_h);
        let t = hwc_to_chw(&resize_hwc(t, resize_w, resize_h)?);
        let t = if resize_w == out_w { t } else { t.f_narrow(2, (resize_w - out_w) / 2, out_w)? };
        let t = if resize_h == out_h { t } else { t.f_narrow(1, (resize_h - out_h) / 2, out_h)? };
        Ok(t)
    }
}

/// Resize an image, preserve the aspect ratio by taking a center crop.
///
/// This expects as input a tensor of shape [channel, height, width] and returns
pub fn resize_preserve_aspect_ratio(
    t: &Tensor,
    out_w: i64,
    out_h: i64,
) -> Result<Tensor, TchError> {
    resize_preserve_aspect_ratio_hwc(&chw_to_hwc(t), out_w, out_h)
}

/// Loads and resize an image, preserve the aspect ratio by taking a center crop.
pub fn load_and_resize<T: AsRef<Path>>(
    path: T,
    out_w: i64,
    out_h: i64,
) -> Result<Tensor, TchError> {
    let tensor = load_hwc(path)?;
    resize_preserve_aspect_ratio_hwc(&tensor, out_w, out_h)
}

/// Loads and resize an image from memory, preserve the aspect ratio by taking a center crop.
pub fn load_and_resize_from_memory(
    img_data: &[u8],
    out_w: i64,
    out_h: i64,
) -> Result<Tensor, TchError> {
    let tensor = load_hwc_from_mem(img_data)?;
    resize_preserve_aspect_ratio_hwc(&tensor, out_w, out_h)
}

fn visit_dirs(dir: &Path, files: &mut Vec<std::fs::DirEntry>) -> Result<(), TchError> {
    if dir.is_dir() {
        for entry in std::fs::read_dir(dir)? {
            let entry = entry?;
            let path = entry.path();
            if path.is_dir() {
                visit_dirs(&path, files)?;
            } else if entry
                .file_name()
                .to_str()
                .is_some_and(|s| s.ends_with(".png") || s.ends_with(".jpg"))
            {
                files.push(entry);
            }
        }
    }
    Ok(())
}

/// Loads all the images in a directory.
pub fn load_dir<T: AsRef<Path>>(path: T, out_w: i64, out_h: i64) -> Result<Tensor, TchError> {
    let mut files: Vec<std::fs::DirEntry> = vec![];
    visit_dirs(path.as_ref(), &mut files)?;
    if files.is_empty() {
        return Err(TchError::Io(io::Error::new(
            io::ErrorKind::NotFound,
            format!("no image found in {:?}", path.as_ref(),),
        )));
    }
    let v: Vec<_> = files
        .iter()
        // Silently discard image errors.
        .filter_map(|x| load_and_resize(x.path(), out_w, out_h).ok())
        .collect();
    Ok(Tensor::stack(&v, 0))
}