jax-ml/jax · error · ValueError

in1 and in2 must have the same number of dimensions

Error message

in1 and in2 must have the same number of dimensions

What it means

_convolve_nd requires in1.ndim == in2.ndim; unlike some frameworks it does not auto-promote ranks for convolution. Note padding semantics differ from jnp.convolve, hence the separate implementation.

Source

Thrown at jax/_src/scipy/signal.py:166

    out_shape = in1.shape
  elif mode == "valid":
    out_shape = tuple(s1 - s2 + 1 for s1, s2 in zip(in1.shape, in2.shape))
  else:
    raise ValueError(f"Unrecognized {mode=}")

  start_indices = tuple((full_size - out_size) // 2
                        for full_size, out_size in zip(full_shape, out_shape))
  return lax.dynamic_slice(conv, start_indices, out_shape)


# Note: we do not reuse the code from jax.numpy.convolve here, because the handling
# of padding differs slightly between the two implementations (particularly for
# mode='same').
def _convolve_nd(in1: Array, in2: Array, mode: ModeString, *, precision: PrecisionLike) -> Array:
  if mode not in ["full", "same", "valid"]:
    raise ValueError("mode must be one of ['full', 'same', 'valid']")
  if in1.ndim != in2.ndim:
    raise ValueError("in1 and in2 must have the same number of dimensions")
  if in1.size == 0 or in2.size == 0:
    raise ValueError(f"zero-size arrays not supported in convolutions, got shapes {in1.shape} and {in2.shape}.")
  in1, in2 = promote_dtypes_inexact(in1, in2)

  no_swap = all(s1 >= s2 for s1, s2 in zip(in1.shape, in2.shape))
  swap = all(s1 <= s2 for s1, s2 in zip(in1.shape, in2.shape))
  if not (no_swap or swap):
    raise ValueError("One input must be smaller than the other in every dimension.")

  shape_o = in2.shape
  if swap:
    in1, in2 = in2, in1
  shape = in2.shape
  in2 = jnp.flip(in2)

  if mode == 'valid':
    padding = [(0, 0) for s in shape]
  elif mode == 'same':

View on GitHub (pinned to 1e1c6a8fc0)

Solutions

Reshape to match ranks: kernel[None, :] for 2D, kernel[None, None] for batched 3D
Use jnp.squeeze on stray unit dims deliberately
Prefer fftconvolve with axes= for selective convolution

Example fix

# before
signal.convolve2d(img, kernel1d, mode='same')
# after
signal.convolve2d(img, kernel1d[None, :], mode='same')

Defensive patterns

Strategy: validation

Validate before calling

assert in1.ndim == in2.ndim, (in1.shape, in2.shape)

Type guard

def same_rank(a, b) -> bool: return jnp.asarray(a).ndim == jnp.asarray(b).ndim

Prevention

Use a helper that reshapes kernels to input rank automatically

When it happens

Trigger: convolve2d expects 2D inputs; passing a (H,W) image with a (K,) kernel, or a batched (N,H,W) with a 2D filter.

Common situations: 1D-vs-2D kernel confusion, batched inputs forgetting the leading dim on the filter.

Related errors

AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/27168e348bf7533e. Report an issue: GitHub.