pub fn preprocess_to_nchw(page: &RgbPage) -> Vec<f32>Expand description
Preprocess a render into the detector’s NCHW f32 input (1x3x640x640).
Resize to 640x640 non-uniform with PIL BILINEAR (triangle support widened on downscale, as PIL does), then rescale by 1/255 with no mean/std normalisation, laid out channel-planar.