Struct Layout

Source

pub struct Layout {
    pub threshold: f32,
    /* private fields */
}

Expand description

载入好的版面模型。

Fields§

§threshold: f32

分数阈值；缺省同参考实现，可按需调整。

Implementations§

Source §

impl Layout

Source

pub fn load(dir: &Path) -> Result<Layout, String>

dir 是模型目录：model.safetensors、config.json、preprocessor_config.json，可选 inference.yml。

Examples found in repository ?

examples/layout_check.rs (line 55)

48fn main() {
49    let a: Vec<String> = std::env::args().skip(1).collect();
50    assert!(
51        a.len() == 3 || a.len() == 4,
52        "用法：layout_check <模型目录> <图> <输出> [分数阈值]"
53    );
54    let t = Instant::now();
55    let mut m = pagesift_ocr::layout::Layout::load(std::path::Path::new(&a[0])).unwrap();
56    if let Some(th) = a.get(3) {
57        m.threshold = th.parse().unwrap();
58    }
59    eprintln!("载入 {:.0} ms", t.elapsed().as_secs_f64() * 1e3);
60    let img = read_pnm(&a[1]);
61    // 设了 LAYOUT_RAW 就打印原始输出的左上角，和参考实现的集成测试对照
62    if std::env::var_os("LAYOUT_RAW").is_some() {
63        let r = m.raw(&img).unwrap();
64        let (_, k) = r.logits.tc();
65        let (q, _) = r.order.tc();
66        for i in 0..3 {
67            println!("logits {:?}", &r.logits.data[i * k..i * k + 3]);
68            println!("boxes {:?}", &r.boxes[i][..3]);
69            println!("order {:?}", &r.order.data[i * q..i * q + 3]);
70        }
71    }
72    let t = Instant::now();
73    let rs = m.detect(&img).unwrap();
74    let ms = t.elapsed().as_secs_f64() * 1e3;
75    eprintln!(
76        "{}: {}×{}，{} 个区域，{ms:.0} ms",
77        a[1],
78        img.width,
79        img.height,
80        rs.len()
81    );
82    let mut s = String::new();
83    for r in &rs {
84        writeln!(
85            s,
86            "{:.4} {:.4} {:.4} {:.4} {:.4} {} {}",
87            r.x0, r.y0, r.x1, r.y1, r.score, r.label, r.kind
88        )
89        .unwrap();
90    }
91    std::fs::write(&a[2], s).unwrap();
92}

Source

pub fn detect(&self, img: &Image) -> Result<Vec<Region>, String>

检测整页图的版面区域，按预测的阅读顺序返回。

Examples found in repository ?

examples/layout_check.rs (line 73)

48fn main() {
49    let a: Vec<String> = std::env::args().skip(1).collect();
50    assert!(
51        a.len() == 3 || a.len() == 4,
52        "用法：layout_check <模型目录> <图> <输出> [分数阈值]"
53    );
54    let t = Instant::now();
55    let mut m = pagesift_ocr::layout::Layout::load(std::path::Path::new(&a[0])).unwrap();
56    if let Some(th) = a.get(3) {
57        m.threshold = th.parse().unwrap();
58    }
59    eprintln!("载入 {:.0} ms", t.elapsed().as_secs_f64() * 1e3);
60    let img = read_pnm(&a[1]);
61    // 设了 LAYOUT_RAW 就打印原始输出的左上角，和参考实现的集成测试对照
62    if std::env::var_os("LAYOUT_RAW").is_some() {
63        let r = m.raw(&img).unwrap();
64        let (_, k) = r.logits.tc();
65        let (q, _) = r.order.tc();
66        for i in 0..3 {
67            println!("logits {:?}", &r.logits.data[i * k..i * k + 3]);
68            println!("boxes {:?}", &r.boxes[i][..3]);
69            println!("order {:?}", &r.order.data[i * q..i * q + 3]);
70        }
71    }
72    let t = Instant::now();
73    let rs = m.detect(&img).unwrap();
74    let ms = t.elapsed().as_secs_f64() * 1e3;
75    eprintln!(
76        "{}: {}×{}，{} 个区域，{ms:.0} ms",
77        a[1],
78        img.width,
79        img.height,
80        rs.len()
81    );
82    let mut s = String::new();
83    for r in &rs {
84        writeln!(
85            s,
86            "{:.4} {:.4} {:.4} {:.4} {:.4} {} {}",
87            r.x0, r.y0, r.x1, r.y1, r.score, r.label, r.kind
88        )
89        .unwrap();
90    }
91    std::fs::write(&a[2], s).unwrap();
92}

Source

pub fn raw(&self, img: &Image) -> Result<Raw, String>

网络的原始输出（后处理之前），用来和参考实现逐数对照。

Examples found in repository ?

examples/layout_check.rs (line 63)

48fn main() {
49    let a: Vec<String> = std::env::args().skip(1).collect();
50    assert!(
51        a.len() == 3 || a.len() == 4,
52        "用法：layout_check <模型目录> <图> <输出> [分数阈值]"
53    );
54    let t = Instant::now();
55    let mut m = pagesift_ocr::layout::Layout::load(std::path::Path::new(&a[0])).unwrap();
56    if let Some(th) = a.get(3) {
57        m.threshold = th.parse().unwrap();
58    }
59    eprintln!("载入 {:.0} ms", t.elapsed().as_secs_f64() * 1e3);
60    let img = read_pnm(&a[1]);
61    // 设了 LAYOUT_RAW 就打印原始输出的左上角，和参考实现的集成测试对照
62    if std::env::var_os("LAYOUT_RAW").is_some() {
63        let r = m.raw(&img).unwrap();
64        let (_, k) = r.logits.tc();
65        let (q, _) = r.order.tc();
66        for i in 0..3 {
67            println!("logits {:?}", &r.logits.data[i * k..i * k + 3]);
68            println!("boxes {:?}", &r.boxes[i][..3]);
69            println!("order {:?}", &r.order.data[i * q..i * q + 3]);
70        }
71    }
72    let t = Instant::now();
73    let rs = m.detect(&img).unwrap();
74    let ms = t.elapsed().as_secs_f64() * 1e3;
75    eprintln!(
76        "{}: {}×{}，{} 个区域，{ms:.0} ms",
77        a[1],
78        img.width,
79        img.height,
80        rs.len()
81    );
82    let mut s = String::new();
83    for r in &rs {
84        writeln!(
85            s,
86            "{:.4} {:.4} {:.4} {:.4} {:.4} {} {}",
87            r.x0, r.y0, r.x1, r.y1, r.score, r.label, r.kind
88        )
89        .unwrap();
90    }
91    std::fs::write(&a[2], s).unwrap();
92}