Datasets:
File size: 2,138 Bytes
b0d8bdd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 | # Copyright 2026 Nikhil Raghavendra
# SPDX-License-Identifier: Apache-2.0
"""Shared fixtures: deterministic structured images and the weights guard."""
from __future__ import annotations
from importlib.util import find_spec
from typing import TYPE_CHECKING
import numpy as np
import pytest
from sen_bridge import weights
if TYPE_CHECKING:
from numpy.typing import NDArray
_SIZE = 128
_BLUR = 5
# Skip modules whose framework is not installed instead of failing collection.
collect_ignore: list[str] = []
if find_spec("torch") is None:
collect_ignore += ["test_metrics.py", "test_parity.py"]
if find_spec("jax") is None:
collect_ignore.append("test_parity.py")
def structured_images(n: int, seed: int = 0) -> NDArray[np.float32]:
"""Make deterministic images in [-1, 1]: smoothed noise plus a gradient.
Args:
n: Number of images.
seed: Seed of the noise.
Returns:
Float32 array of shape ``[n, 3, 128, 128]``.
"""
rng = np.random.default_rng(seed)
a = rng.random((n, 3, _SIZE, _SIZE)).astype(np.float32)
kernel = np.ones(_BLUR, np.float32) / _BLUR
for axis in (2, 3):
a = np.apply_along_axis(
lambda v: np.convolve(v, kernel, mode="same"), axis, a
)
g = np.linspace(-1, 1, _SIZE, dtype=np.float32)
ramp = 0.5 * g[None, None, :, None] * g[None, None, None, :]
return np.clip(a * 3 - 1 + ramp, -1, 1).astype(np.float32)
@pytest.fixture(scope="session")
def pair() -> tuple[NDArray[np.float32], NDArray[np.float32]]:
"""Return a batch of images and a noisy copy of it."""
x = structured_images(12)
noise = np.random.default_rng(1).standard_normal(x.shape).astype(np.float32)
return x, np.clip(x + 0.15 * noise, -1, 1).astype(np.float32)
@pytest.fixture(scope="session")
def weights_ready() -> None:
"""Make sure the metric weights are available, or skip the test."""
try:
weights.path("fid_inception")
weights.path("lpips_vgg")
# Offline or the host is unavailable.
except Exception as exc: # ruff: ignore[blind-except]
pytest.skip(f"weights unavailable: {exc}")
|