Struct DBPostProcess 

pub struct DBPostProcess {
    pub thresh: f32,
    pub box_thresh: f32,
    pub max_candidates: usize,
    pub unclip_ratio: f32,
    pub min_size: f32,
    pub score_mode: ScoreMode,
    pub box_type: BoxType,
    pub use_dilation: bool,
}

Post-processor for DB (Differentiable Binarization) text detection models.

Fields

thresh: f32

Threshold for binarizing the prediction map (default: 0.3).

box_thresh: f32

Threshold for filtering bounding boxes based on their score (default: 0.7).

max_candidates: usize

Maximum number of candidate bounding boxes to consider (default: 1000).

unclip_ratio: f32

Ratio for unclipping (expanding) bounding boxes (default: 2.0).

min_size: f32

Minimum side length for detected bounding boxes.

score_mode: ScoreMode

Method for calculating the score of a bounding box.

box_type: BoxType

Type of bounding box to generate (quadrilateral or polygon).

use_dilation: bool

Whether to apply dilation to the segmentation mask before contour detection.

Implementations

impl DBPostProcess

pub fn box_score_fast(&self, pred: &Array2<f32>, bbox: &BoundingBox) -> f32

Calculates the score of a bounding box using a fast approximation method.

impl DBPostProcess

pub fn new( thresh: Option<f32>, box_thresh: Option<f32>, max_candidates: Option<usize>, unclip_ratio: Option<f32>, use_dilation: Option<bool>, score_mode: Option<ScoreMode>, box_type: Option<BoxType>, ) -> Self

Creates a new DBPostProcess instance with optional overrides.

pub fn apply( &self, preds: &Tensor4D, img_shapes: Vec<[f32; 4]>, thresh: Option<f32>, box_thresh: Option<f32>, unclip_ratio: Option<f32>, ) -> (Vec<Vec<BoundingBox>>, Vec<Vec<f32>>)

Applies post-processing to a batch of prediction maps.

Trait Implementations

impl Debug for DBPostProcess

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

Formats the value using the given formatter.