HappyOnigiri
4d45ea6e25
- Rename py/ directory to meld/ to eliminate PyPI 'py' package shadowing, which required conftest.py/path.py workarounds in the test suite - Remove meld/path.py (pytest stub no longer needed after rename) - Empty meld/__init__.py (removed redundant 'from . import path as path') - Replace 87-line conftest.py monkeypatch with minimal Package.setup() shim - Extract create_schema(cursor) from schema.py as single DDL source of truth, called by both init_db() and test helpers - Add permanent_delete, soft_delete, soft_delete_to_trash, restore_image, restore_image_with_rename, and collect_deleted_ancestors to repository.py to eliminate duplicated inline SQL across router.py and importer/service.py - Update all import paths, pyproject.toml, Makefile, CI config, and check_scripts to reference meld/ instead of py/
97 lines
3.7 KiB
Python
97 lines
3.7 KiB
Python
import os
|
|
import sys
|
|
import unittest
|
|
|
|
import torch
|
|
|
|
# Add project root to path to enable imports
|
|
sys.path.append(os.path.dirname(os.path.dirname(__file__)))
|
|
|
|
from meld.pixelate.nodes import MeldPixelate
|
|
|
|
|
|
class TestPixelate(unittest.TestCase):
|
|
def setUp(self) -> None:
|
|
self.node = MeldPixelate()
|
|
|
|
def test_node_interface(self) -> None:
|
|
"""Confirm that the node interface matches ComfyUI's expectations"""
|
|
self.assertTrue(hasattr(MeldPixelate, "INPUT_TYPES"))
|
|
self.assertTrue(hasattr(MeldPixelate, "RETURN_TYPES"))
|
|
self.assertTrue(hasattr(MeldPixelate, "FUNCTION"))
|
|
self.assertTrue(hasattr(MeldPixelate, "CATEGORY"))
|
|
|
|
inputs = MeldPixelate.INPUT_TYPES()
|
|
self.assertIn("required", inputs)
|
|
self.assertIn("image", inputs["required"])
|
|
self.assertIn("pixel_size", inputs["required"])
|
|
|
|
self.assertEqual(MeldPixelate.RETURN_TYPES, ("IMAGE",))
|
|
self.assertEqual(MeldPixelate.FUNCTION, "pixelate")
|
|
|
|
def test_pixelate_basic(self) -> None:
|
|
"""Confirm that basic pixelation works correctly and maintains the original size"""
|
|
# [Batch, Height, Width, Channels]
|
|
dummy_image = torch.zeros((1, 64, 64, 3))
|
|
# Create a single white point in the center
|
|
dummy_image[0, 32, 32, :] = 1.0
|
|
|
|
pixel_size = 8
|
|
(result,) = self.node.pixelate(dummy_image, pixel_size)
|
|
|
|
# Confirm that the output size is the same as the input size
|
|
self.assertEqual(result.shape, dummy_image.shape)
|
|
self.assertIsInstance(result, torch.Tensor)
|
|
|
|
# Confirm that the white point at 32,32 has expanded to an 8x8 block due to pixelation
|
|
# 32//8 = 4, so check the block range
|
|
block_start_h = (32 // pixel_size) * pixel_size
|
|
block_end_h = block_start_h + pixel_size
|
|
block_start_w = (32 // pixel_size) * pixel_size
|
|
block_end_w = block_start_w + pixel_size
|
|
|
|
# Confirm that values inside the block are greater than 0 (white point is dispersed)
|
|
# Due to 'area' interpolation, a value of 1/64 (for scale=8) should spread within the block
|
|
self.assertTrue(torch.all(result[0, block_start_h:block_end_h, block_start_w:block_end_w, :] > 0))
|
|
# Confirm that areas outside the block are 0
|
|
self.assertTrue(torch.all(result[0, 0:block_start_h, :, :] == 0))
|
|
self.assertTrue(torch.all(result[0, block_end_h:, :, :] == 0))
|
|
|
|
def test_pixelate_batch(self) -> None:
|
|
"""Confirm that batch processing is handled correctly"""
|
|
batch_size = 2
|
|
dummy_images = torch.rand((batch_size, 32, 32, 3))
|
|
pixel_size = 4
|
|
|
|
(result,) = self.node.pixelate(dummy_images, pixel_size)
|
|
|
|
self.assertEqual(result.shape, (batch_size, 32, 32, 3))
|
|
|
|
def test_pixelate_min_scale(self) -> None:
|
|
"""Confirm that there is no substantial change when pixel_size=1 (due to nearest-exact properties)"""
|
|
dummy_image = torch.rand((1, 16, 16, 3))
|
|
pixel_size = 1
|
|
|
|
(result,) = self.node.pixelate(dummy_image, pixel_size)
|
|
|
|
# Should be identical if scale=1
|
|
self.assertTrue(torch.allclose(dummy_image, result))
|
|
|
|
def test_pixelate_large_scale(self) -> None:
|
|
"""Confirm that it results in 1x1 without error even if pixel_size is larger than the image size"""
|
|
dummy_image = torch.rand((1, 8, 8, 3))
|
|
pixel_size = 16 # Larger than image size
|
|
|
|
(result,) = self.node.pixelate(dummy_image, pixel_size)
|
|
|
|
self.assertEqual(result.shape, (1, 8, 8, 3))
|
|
# All pixels should have the same value (average)
|
|
first_pixel = result[0, 0, 0, :]
|
|
for h in range(8):
|
|
for w in range(8):
|
|
self.assertTrue(torch.allclose(result[0, h, w, :], first_pixel))
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|