Struct Config 

pub struct Config {
    pub stem_channels: [usize; 3],
    pub blocks: Vec<Vec<BlockConfig>>,
}

骨干的配置（config.json 里的 backbone_config）。

Fields

stem_channels: [usize; 3]

blocks: Vec<Vec<BlockConfig>>

Implementations

impl Config

pub fn from_json(v: &Value) -> Result<Config, String>

从 config.json 的 backbone_config 读。只支持 stem_type = "large"、缩放 1.0、 不用可学习仿射块（PP-OCRv6 的检测、识别都是这样）。

Examples found in repository

examples/backbone_smoke.rs (line 13)

fn main() {
    let a: Vec<String> = std::env::args().skip(1).collect();
    let dir = std::path::Path::new(&a[0]);
    let (h, w): (usize, usize) = (a[1].parse().unwrap(), a[2].parse().unwrap());
    let cfg: serde_json::Value =
        serde_json::from_slice(&std::fs::read(dir.join("config.json")).unwrap()).unwrap();
    let wts =
        pagesift_infer::Weights::from_bytes(&std::fs::read(dir.join("model.safetensors")).unwrap())
            .unwrap();
    let c = pagesift_ocr::lcnet::Config::from_json(&cfg).unwrap();
    let b = pagesift_ocr::lcnet::Backbone::load(&wts, "model.backbone.encoder", &c).unwrap();
    let x = pagesift_infer::Tensor::new(
        vec![3, h, w],
        (0..3 * h * w)
            .map(|i| ((i % 97) as f32 / 97.0) - 0.5)
            .collect(),
    );
    let t = Instant::now();
    let outs = b.forward(&x);
    for o in &outs {
        println!(
            "{:?} 均值 {:.5}",
            o.shape,
            o.data.iter().sum::<f32>() / o.data.len() as f32
        );
    }
    println!("{:.1} ms", t.elapsed().as_secs_f64() * 1e3);
}

Trait Implementations

impl Clone for Config

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Config

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.