Skip to main content

conv_bn

vision_rs::models::yolo::yolo26::blocks::conv

Function conv_bn 

Source 
pub fn conv_bn<D: Float>(
    c_in: usize,
    c_out: usize,
    k: usize,
    s: usize,
    g: usize,
) -> impl Fn(SymTensor) -> SymTensor
Expand description

Conv2d(groups) → BatchNorm2d — no activation.

Used for layers in PSABlock that skip the SiLU activation (qkv, pe, proj, ffn1). g=1 for standard 1×1 convs; g=c_in=c_out for depthwise. BN eps=0.001 matching the pretrained YOLO26n weights.