Struct Topk 

pub struct Topk { /* private fields */ }

A processor for extracting top-k results from classification outputs.

The Topk struct processes classification model outputs to extract the top-k most confident predictions along with their class names (if available).

Implementations

impl Topk

pub fn new(class_id_map: Option<HashMap<usize, String>>) -> Self

Creates a new Topk processor with optional class name mapping.

Arguments

• class_id_map - Optional mapping from class IDs to human-readable class names.

Examples

use std::collections::HashMap;
use oar_ocr::utils::topk::Topk;

let mut class_map = HashMap::new();
class_map.insert(0, "cat".to_string());
class_map.insert(1, "dog".to_string());

let topk = Topk::new(Some(class_map));

pub fn without_class_names() -> Self

Creates a new Topk processor without class name mapping.

Examples

use oar_ocr::utils::topk::Topk;

let topk = Topk::without_class_names();

pub fn from_class_names(class_names: Vec<String>) -> Self

Creates a new Topk processor with class names from a vector.

The vector index corresponds to the class ID.

Arguments

• class_names - Vector of class names where index = class ID.

Examples

use oar_ocr::utils::topk::Topk;

let class_names = vec!["cat".to_string(), "dog".to_string(), "bird".to_string()];
let topk = Topk::from_class_names(class_names);

pub fn process( &self, predictions: &[Vec<f32>], k: usize, ) -> Result<TopkResult, String>

Processes classification outputs to extract top-k results.

Arguments

• predictions - 2D vector where each inner vector contains the confidence scores for all classes for one prediction.
• k - Number of top predictions to extract (must be > 0).

Returns

• Ok(TopkResult) - The top-k results with indexes, scores, and optional class names.
• Err(String) - If k is 0 or if the input is invalid.

Examples

use oar_ocr::utils::topk::Topk;

let topk = Topk::without_class_names();
let predictions = vec![
    vec![0.1, 0.8, 0.1],  // Prediction 1: class 1 has highest score
    vec![0.7, 0.2, 0.1],  // Prediction 2: class 0 has highest score
];
let result = topk.process(&predictions, 2).unwrap();

pub fn get_class_name(&self, class_id: usize) -> Option<&String>

Gets the class name for a given class ID.

Arguments

• class_id - The class ID to look up.

Returns

• Option<&String> - The class name if available.

pub fn has_class_names(&self) -> bool

Checks if class name mapping is available.

Returns

• true - If class name mapping is available.
• false - If no class name mapping is available.

pub fn num_classes(&self) -> Option<usize>

Gets the number of classes in the mapping.

Returns

• Option<usize> - Number of classes if mapping is available.

pub fn set_class_mapping( &mut self, class_id_map: Option<HashMap<usize, String>>, )

Updates the class name mapping.

Arguments

• class_id_map - New class ID to name mapping.

pub fn process_single( &self, prediction: &[f32], k: usize, ) -> Result<TopkResult, String>

Processes a single prediction vector.

Arguments

• prediction - Vector of confidence scores for all classes.
• k - Number of top predictions to extract.

Returns

• Ok(TopkResult) - The top-k results for the single prediction.
• Err(String) - If k is 0 or if the input is invalid.

Trait Implementations

impl Debug for Topk

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

Formats the value using the given formatter.

impl Default for Topk

fn default() -> Self

Creates a default Topk processor without class name mapping.