Question 1

How is output size calculated for Conv2D?

Accepted Answer

Output height = floor((H + 2*padH - dilationH*(kH-1) - 1)/strideH + 1). Same for width. Output channels = number of filters.

Question 2

What is the formula for TransposedConv2D (deconv)?

Accepted Answer

Output = (input-1)*stride - 2*padding + dilation*(kernel-1) + output_padding + 1. Output padding is supported (default 0).

Question 3

How are FLOPs estimated?

Accepted Answer

FLOPs = 2 × kH × kW × C_in × out_H × out_W × C_out (multiply-add counts as 2 FLOPs). Bias addition is negligible.

🧠 CNN Output Size Calculator ml3x

🧩 Multi-layer network