Image Pipeline Playground
"""Tests for Albumentations multi-target augment nodes."""
from __future__ import annotations
import numpy as np
from app.engine.cache import CacheManager
from app.engine.executor import DagExecutor
from app.engine.registry import registry
from app.models.graph import Edge, ExecuteRequest, Graph, NodeInstance
from app.nodes import register_builtin_nodes
def _image() -> np.ndarray:
image = np.full((64, 64, 3), 90, dtype=np.uint8)
image[8:40, 8:40] = (30, 180, 70)
return image
def test_augment_nodes_registered() -> None:
register_builtin_nodes()
augment = [meta for meta in registry.list_metadata() if meta.category == "augment"]
assert len(augment) >= 50
assert registry.has("albu_horizontal_flip")
assert registry.has("albu_random_brightness_contrast")
assert registry.has("annotations")
def test_image_only_passthrough_annotations() -> None:
register_builtin_nodes()
node = registry.get("albu_random_brightness_contrast")
params = node.validate_params({"p": 1.0, "brightness_limit": 0.1, "contrast_limit": 0.1})
bboxes = [[8, 8, 40, 40, "car"]]
keypoints = [[20.0, 20.0]]
mask = np.zeros((64, 64), dtype=np.uint8)
mask[8:40, 8:40] = 255
out = node.execute(
{"image": _image(), "mask": mask, "bboxes": bboxes, "keypoints": keypoints},
params,
seed=3,
)
assert out["image"].shape == (64, 64, 3)
assert out["bboxes"] == bboxes
assert out["keypoints"] == keypoints
assert out["mask"].shape[:2] == (64, 64)
def test_dual_flip_updates_bboxes_and_keypoints() -> None:
register_builtin_nodes()
node = registry.get("albu_horizontal_flip")
params = node.validate_params({"p": 1.0})
out = node.execute(
{
"image": _image(),
"bboxes": [[8, 8, 40, 40, "car"]],
"keypoints": [[20.0, 12.0]],
},
params,
seed=0,
)
# x mirrored in 64-wide image: x' = W - x
assert out["bboxes"][0][0] == 24.0
assert out["bboxes"][0][2] == 56.0
assert out["bboxes"][0][4] == "car"
assert out["keypoints"][0][0] == 43.0
assert out["keypoints"][0][1] == 12.0
def test_seed_reproducibility() -> None:
register_builtin_nodes()
node = registry.get("albu_gauss_noise")
params = node.validate_params({"p": 1.0})
a = node.execute({"image": _image()}, params, seed=11)["image"]
b = node.execute({"image": _image()}, params, seed=11)["image"]
c = node.execute({"image": _image()}, params, seed=12)["image"]
assert np.array_equal(a, b)
assert not np.array_equal(a, c)
def test_optional_ports_omitted() -> None:
register_builtin_nodes()
node = registry.get("albu_gaussian_blur")
params = node.validate_params({"p": 1.0, "blur_limit": 5})
out = node.execute({"image": _image()}, params, seed=0)
assert set(out.keys()) == {"image"}
def test_annotations_node_and_pipeline(tmp_path) -> None:
register_builtin_nodes()
ann = registry.get("annotations")
payload = ann.execute(
{},
ann.validate_params(
{
"bboxes_json": '[[5, 5, 30, 30, "a"]]',
"keypoints_json": "[[10, 12]]",
}
),
seed=0,
)
assert payload["bboxes"][0][4] == "a"
assert payload["keypoints"] == [[10.0, 12.0]]
# Cache mixed ports from a dual node.
flip = registry.get("albu_horizontal_flip")
produced = flip.execute(
{"image": _image(), "bboxes": payload["bboxes"], "keypoints": payload["keypoints"]},
flip.validate_params({"p": 1.0}),
seed=0,
)
cache = CacheManager(tmp_path / "cache")
key = cache.make_key(
"albu_horizontal_flip",
{"p": 1.0},
{port: cache.hash_value(value) for port, value in produced.items()},
seed=0,
)
cache.put_outputs(key, produced)
loaded = cache.get_outputs(key)
assert loaded is not None
assert loaded["bboxes"] == produced["bboxes"]
assert np.array_equal(loaded["image"], produced["image"])
def test_executor_runs_annotations_to_flip(tmp_path) -> None:
register_builtin_nodes()
image_path = tmp_path / "in.png"
import cv2
cv2.imwrite(str(image_path), _image())
request = ExecuteRequest(
graph=Graph(
nodes=[
NodeInstance(
id="load",
type="load_image",
params={"path": str(image_path)},
),
NodeInstance(
id="ann",
type="annotations",
params={
"bboxes_json": "[[8, 8, 40, 40, \"car\"]]",
"keypoints_json": "[[20, 20]]",
},
),
NodeInstance(id="flip", type="albu_horizontal_flip", params={"p": 1.0}),
],
edges=[
Edge(
id="e1",
source="load",
source_port="image",
target="flip",
target_port="image",
),
Edge(
id="e2",
source="ann",
source_port="bboxes",
target="flip",
target_port="bboxes",
),
Edge(
id="e3",
source="ann",
source_port="keypoints",
target="flip",
target_port="keypoints",
),
],
),
seed=0,
cache=True,
)
events: list[dict] = []
executor = DagExecutor(tmp_path / "cache", node_registry=registry)
result = executor.execute(
request,
on_event=lambda event: events.append(event.model_dump())
if event.type.value == "preview"
else None,
)
assert "flip" in result["order"]
flip_previews = [
event
for event in events
if event.get("node_id") == "flip" and event.get("port_id") == "image"
]
assert flip_previews
assert flip_previews[0].get("data") is not None
assert "bboxes" in flip_previews[0]["data"]