jax-ml/jax · error · ValueError

zero-size arrays not supported in convolutions, got shapes {

Error message

zero-size arrays not supported in convolutions, got shapes {in1.shape} and {in2.shape}.

What it means

Raised by jax.scipy.signal._convolve_nd when either input array to a convolution/correlation has zero elements (size 0). JAX's convolution implementation is backed by lax.conv_general_dilated, which cannot handle empty tensors, unlike some SciPy paths. The message reports both input shapes so you can see which operand is empty.

Source

Thrown at jax/_src/scipy/signal.py:168

    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':
    padding = [(s - 1 - (s_o - 1) // 2, s - s_o + (s_o - 1) // 2)
               for (s, s_o) in zip(shape, shape_o)]

View on GitHub (pinned to 1e1c6a8fc0)

Solutions

1. Check in1.size > 0 and in2.size > 0 before calling and skip/handle empty batches explicitly
2. Fix upstream slicing/masking logic that produced the zero-size array
3. If an empty result is semantically valid, return a correctly-shaped zero array instead of calling convolve

Example fix

// before
out = jax.scipy.signal.convolve(x[mask], kernel, mode='same')
// after
sub = x[mask]
out = jax.scipy.signal.convolve(sub, kernel, mode='same') if sub.size else jnp.zeros_like(sub)

def safe_conv(in1, in2, mode='full'):
    if in1.size == 0 or in2.size == 0:
        return None  # or a zero array of the expected output shape
    return jax.scipy.signal.convolve(in1, in2, mode=mode)

def is_nonempty(arr) -> bool:
    return jnp.asarray(arr).size > 0

When it happens

Trigger: Calling jax.scipy.signal.convolve, convolve2d, correlate, or correlate2d with an empty array, e.g. jnp.zeros((0, 5)) or jnp.array([]), or with a shape containing a 0 dimension produced by slicing/boolean masking.

Common situations: Dynamic slicing or filtering that occasionally yields zero rows before convolution; edge cases in batch processing where a batch is empty; empty kernels constructed from data-driven sizes.

Related errors

• convolve2d() only supports 2-dimensional inputs.
• conv_general_dilated batch_group_count must divide lhs batch
• conv_general_dilated rhs output feature dimension size must
• conv_general_dilated window and window_strides must have the
• Wrong number of explicit pads for convolution: expected {},

AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/a81528476ce65024. Report an issue: GitHub.