Æmotion StudioandClaude Fable 5.1
c487613832
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
85 lines
3.2 KiB
Python
85 lines
3.2 KiB
Python
"""
|
|
The interference generator: fringes, and the knobs that set them.
|
|
"""
|
|
import pytest
|
|
import torch
|
|
|
|
from snk.core.constants import CHANNEL_BASIS
|
|
from snk.core.shader_noise import effective_channel_rank, generate
|
|
|
|
CPU = torch.device("cpu")
|
|
PARAMS = {
|
|
"scale": 1.0, "octaves": 3.0, "warp_strength": 0.5, "phase_shift": 0.5,
|
|
"shape_type": "none", "color_scheme": "none", "time": 0.0, "base_seed": 8888,
|
|
}
|
|
|
|
|
|
def _draw(**overrides):
|
|
return generate((1, 8, 64, 64), dict(PARAMS, **overrides), "interference", 8888, CPU)
|
|
|
|
|
|
def _zero_crossings_per_row(noise):
|
|
signs = noise.sign()
|
|
return (signs[..., :, 1:] != signs[..., :, :-1]).float().mean().item()
|
|
|
|
|
|
def _lag_correlation(noise, lag):
|
|
pairs = torch.stack([noise[..., :, :-lag].flatten(), noise[..., :, lag:].flatten()])
|
|
return torch.corrcoef(pairs)[0, 1].item()
|
|
|
|
|
|
def test_warp_strength_is_the_fringe_density():
|
|
soft, mid, dense = (_zero_crossings_per_row(_draw(warp_strength=w)) for w in (0.0, 2.0, 5.0))
|
|
assert soft < mid < dense, (soft, mid, dense)
|
|
|
|
|
|
def test_the_fringes_are_finer_than_the_field_but_still_large_scale():
|
|
"""
|
|
Cutting fringes into an FBM adds structure below the FBM's own scale, which is
|
|
the point, but the default must stay coarse enough for a model to read: with
|
|
the gain set twice as high the field was white within four pixels.
|
|
"""
|
|
fringes = _draw()
|
|
fbm = generate((1, 8, 64, 64), PARAMS, "fractal", 8888, CPU)
|
|
assert _zero_crossings_per_row(fringes) > _zero_crossings_per_row(fbm)
|
|
assert _lag_correlation(fringes, 4) > 0.3
|
|
|
|
|
|
def test_phase_shift_retunes_the_second_field():
|
|
assert not torch.equal(_draw(phase_shift=0.0), _draw(phase_shift=2.0))
|
|
|
|
|
|
def test_noise_scale_zooms():
|
|
assert _zero_crossings_per_row(_draw(scale=0.3)) < _zero_crossings_per_row(_draw(scale=3.0))
|
|
|
|
|
|
@pytest.mark.parametrize("shape", [(1, 4, 64, 64), (1, 24, 22, 38), (1, 24, 5, 16, 16),
|
|
(1, 128, 3, 8, 8), (1, 128, 3, 32, 32)])
|
|
def test_every_channel_is_its_own_field(shape):
|
|
rank = effective_channel_rank(generate(shape, PARAMS, "interference", 8888, CPU))
|
|
assert rank > min(shape[1], CHANNEL_BASIS) * 0.6, rank
|
|
|
|
|
|
def test_channel_zero_does_not_depend_on_how_many_channels_were_asked_for():
|
|
wide = generate((1, 24, 5, 22, 38), PARAMS, "interference", 8888, CPU)
|
|
single = generate((1, 1, 5, 22, 38), PARAMS, "interference", 8888, CPU)
|
|
assert torch.equal(wide[:, :1], single)
|
|
|
|
|
|
@pytest.mark.parametrize("channel", [1, 5])
|
|
def test_a_batched_channel_is_the_draw_its_seed_gives_alone(channel):
|
|
wide = generate((1, 8, 22, 39), PARAMS, "interference", 8888, CPU)
|
|
alone = generate((1, 1, 22, 39), PARAMS, "interference", 8888 + 6151 * channel, CPU)
|
|
assert torch.equal(wide[:, channel:channel + 1], alone)
|
|
|
|
|
|
def test_holding_the_seed_and_advancing_time_evolves_one_field():
|
|
def frame_correlation(coherent):
|
|
noise = generate((1, 24, 5, 22, 38), dict(PARAMS, use_temporal_coherence=coherent),
|
|
"interference", 8888, CPU, temporal_coherence=coherent)
|
|
pairs = torch.stack([noise[0, :, 0].flatten(), noise[0, :, 1].flatten()])
|
|
return torch.corrcoef(pairs)[0, 1].item()
|
|
|
|
assert frame_correlation(True) > 0.4
|
|
assert abs(frame_correlation(False)) < 0.15
|