Struct CTCLabelDecode 

pub struct CTCLabelDecode { /* private fields */ }

A decoder for CTC (Connectionist Temporal Classification) based text recognition models.

This struct extends BaseRecLabelDecode to provide specialized decoding for CTC models, which include a blank token that needs to be handled specially during decoding.

Fields

• base - The base decoder that handles character mapping and basic decoding operations
• blank_index - The index of the blank token in the character vocabulary

Implementations

impl CTCLabelDecode

pub fn new(character_list: Option<&str>, use_space_char: bool) -> Self

Creates a new CTCLabelDecode instance.

Arguments

• character_list - An optional string containing the character vocabulary. If None, a default alphanumeric character set is used.
• use_space_char - Whether to include a space character in the vocabulary.

Returns

A new CTCLabelDecode instance.

pub fn from_string_list( character_list: Option<&[String]>, use_space_char: bool, has_explicit_blank: bool, ) -> Self

Creates a new CTCLabelDecode instance from a list of strings.

Arguments

• character_list - An optional slice of strings containing the character vocabulary. Only the first character of each string is used. If None, a default alphanumeric character set is used.
• use_space_char - Whether to include a space character in the vocabulary.
• has_explicit_blank - Whether the character list already includes a blank token.

Returns

A new CTCLabelDecode instance.

pub fn get_blank_index(&self) -> usize

Gets the index of the blank token.

Returns

The index of the blank token.

pub fn get_character_list(&self) -> &[char]

Gets the character list used by this decoder.

Returns

A slice containing the characters in the vocabulary.

pub fn get_character_count(&self) -> usize

Gets the number of characters in the vocabulary.

Returns

The number of characters in the vocabulary.

pub fn apply(&self, pred: &Tensor3D) -> (Vec<String>, Vec<f32>)

Applies the CTC decoder to a tensor of model predictions.

This method handles the special requirements of CTC decoding:

1. Removing blank tokens
2. Removing consecutive duplicate characters
3. Converting indices to characters
4. Calculating confidence scores

Arguments

• pred - A 3D tensor containing the model predictions.

Returns

A tuple containing:

• A vector of decoded text strings
• A vector of confidence scores for each text string

Trait Implementations

impl Debug for CTCLabelDecode

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

Formats the value using the given formatter.