Struct RTDetrModel 

pub struct RTDetrModel { /* private fields */ }

RT-DETR layout detection model.

This is a pure model implementation that handles:

• Preprocessing: Image resizing and normalization
• Inference: Running the ONNX model
• Postprocessing: Returning raw predictions

The model is independent of any specific task or adapter.

Implementations

impl RTDetrModel

pub fn new( inference: OrtInfer, preprocess_config: RTDetrPreprocessConfig, ) -> Result<RTDetrModel, OCRError>

Creates a new RT-DETR model.

pub fn preprocess( &self, images: Vec<ImageBuffer<Rgb<u8>, Vec<u8>>>, ) -> Result<(ArrayBase<OwnedRepr<f32>, Dim<[usize; 4]>>, Vec<ImageScaleInfo>, Vec<[f32; 2]>, Vec<[f32; 2]>), OCRError>

Preprocesses images for RT-DETR model.

Returns:

• Batch tensor ready for inference
• Image shapes after resizing [h, w, ratio_h, ratio_w]
• Original shapes [h, w]
• Resized shapes [h, w]

pub fn infer( &self, batch_tensor: &ArrayBase<OwnedRepr<f32>, Dim<[usize; 4]>>, scale_factor: ArrayBase<OwnedRepr<f32>, Dim<[usize; 2]>>, im_shape: ArrayBase<OwnedRepr<f32>, Dim<[usize; 2]>>, ) -> Result<ArrayBase<OwnedRepr<f32>, Dim<[usize; 4]>>, OCRError>

Runs inference on the preprocessed batch tensor.

RT-DETR requires both scale_factor and im_shape inputs.

pub fn postprocess( &self, predictions: ArrayBase<OwnedRepr<f32>, Dim<[usize; 4]>>, _config: &RTDetrPostprocessConfig, ) -> Result<RTDetrModelOutput, OCRError>

Postprocesses model predictions.

For RT-DETR, we just return the raw predictions. The adapter layer will handle converting these to task-specific outputs.

pub fn forward( &self, images: Vec<ImageBuffer<Rgb<u8>, Vec<u8>>>, config: &RTDetrPostprocessConfig, ) -> Result<(RTDetrModelOutput, Vec<ImageScaleInfo>), OCRError>

Runs the complete forward pass: preprocess -> infer -> postprocess.

Trait Implementations

impl Debug for RTDetrModel

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

Formats the value using the given formatter.