codec / predictor · 2D / lossless

wp_static

One learned predictor, fixed for the whole tile.

wp_static looks at W, N, NW, NE and NN. It learns four weights once per tile, stores them in the header and then keeps them fixed, so decoding remains predictable and efficient. With the default weights it behaves like planar. Choose it when a fixed learned model works better than the simpler predictors on your data.

ctid

0x72D707

header

uint32 width · uint8 shift · 4×int16 coeffs
+ uint32 · plane count, when above one

widths

8 · 16 · 32 · 64-bit integers

decode

P = W + ((Σ cᵢ·tapᵢ + round) >> shift), prefix sum per row