torch.Conv2dShape

export type Conv2dShape<S extends Shape, OutChannels extends number> = S extends readonly [
  infer B extends number,
  number,
  number,
  number,
]
  ? readonly [B, OutChannels, number, number]
  : S extends readonly [number, number, number]
    ? readonly [OutChannels, number, number]
    : DynamicShape;

Sextends Shape

OutChannelsextends number

Output shape for 2D convolution. Input: [B, Cin, H, W], Output: [B, Cout, outH, outW] Channel dimension changes to out_channels, spatial dims are dynamic.

Examples

type Input = readonly [16, 3, 224, 224];      // [B, Cin, H, W]
type Output = Conv2dShape<Input, 64>;         // [16, 64, number, number]