File size: 12,717 Bytes
e6f80f3 | 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 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 | """Regression tests for minimal-condition data, not legacy semantic-front labels."""
from copy import deepcopy
import json
import math
import pytest
from fastfill.v2.direct_layout import request_to_condition
from fastfill.v2.batch import TinyTokenizer, collate_samples
from fastfill.v2.io import fingerprint
from fastfill.v2.minimal_data import build_minimal_dataset, project_sample
def pair(uid="room", split="train", floor=1., height=3.):
target = {"id": "obj_0000", "target_size_local_m": [1., .5, 1.],
"bottom_center_m": [11., 21., floor + .5], "yaw_rad": 0.}
room = {"frame": "right_handed_z_up", "room_type": "office", "boundary_known": True,
"floor_known": True, "floor_z_m": floor, "height_m": height,
"floor_polygon_xy_m": [[10., 20.], [14., 20.], [14., 23.], [10., 23.]],
"fixed_objects": [{"id": "fixed_0", "category": "window", "size_local_m": [1., .1, 1.],
"bottom_center_m": [10., 21., floor + 1.], "yaw_rad": 0.}]}
condition = {"schema_version": "fastfill.v2", "room": room,
"objects": [{"id": "obj_0000", "category": "chair", "description": "chair",
"support_parent": "floor", "exchangeable_group": "anonymous_0"}],
"constraints": [{"type": "against_wall", "object_id": "obj_0000"}]}
provenance = {"source": "SpatialLM", "scene_id": uid, "legacy_uid": uid,
"house_id": "house_" + uid, "group": "house_" + uid, "split": split,
"target_source_ids": ["source_0"], "field_evidence": [{"tilted": False,
"size_semantics": "canonical_source_IR", "front_policy": "strict"}]}
sample = {"schema_version": "fastfill.v2", "condition": condition,
"target": {"schema_version": "fastfill.v2", "objects": [target]},
"validity": {"position": [[True] * 3], "size": [[True] * 3], "yaw": [False]},
"provenance": provenance}
source = {"uid": uid, "source": "SpatialLM", "objects": [{"id": "source_0", "size": [1., .5, 1.],
"pos": [11., 21., floor + .5], "yaw": 0., "tilted": False}],
"boundary": room["floor_polygon_xy_m"], "group": provenance["group"],
"room_type": "office", "height": height, "floor_z_m": floor, "boundary_type": "polygon"}
return sample, source
def test_projection_uses_same_public_adapter_and_preserves_lineage():
sample, source = pair()
before = deepcopy(sample)
result = project_sample(sample, source, split="train")
request = {"room_type": "office", "room_size_m": [4., 3.],
"furniture_list": [{"id": "obj_0000", "category": "chair", "description": "chair", "count": 1}]}
assert result["condition"] == request_to_condition(request)
assert result["provenance"]["minimal_request"] == request
assert result["target"]["objects"][0]["bottom_center_m"] == [1., 1., .5]
assert result["target"]["objects"][0]["target_size_local_m"] == [1., .5, 1.]
assert result["provenance"]["frame_translation_m"] == [-10., -20., -1.]
for key in ("scene_id", "legacy_uid", "house_id", "group", "split", "target_source_ids"):
assert result["provenance"][key] == sample["provenance"][key]
assert sample == before
batch = collate_samples([result], TinyTokenizer())
assert batch["scale"][0, 2].item() == 3.
assert result["condition"]["room"]["height_m"] is None
assert result["provenance"]["qualification_source_height_m"] == 3.
def test_xyz_height_is_explicit_opt_in_not_default():
sample, source = pair(height=2.5)
xy = project_sample(sample, source, split="train")
xyz = project_sample(sample, source, split="train", room_dimension_mode="xyz")
assert xy["provenance"]["minimal_request"]["room_size_m"] == [4., 3.]
assert xyz["provenance"]["minimal_request"]["room_size_m"] == [4., 3., 2.5]
assert xyz["condition"]["room"]["height_m"] == 2.5
assert xy["target"] == xyz["target"]
assert collate_samples([xy], TinyTokenizer())["scale"][0, 2].item() == 3.
assert collate_samples([xyz], TinyTokenizer())["scale"][0, 2].item() == 2.5
with pytest.raises(ValueError, match="room_dimension_mode"):
project_sample(sample, source, split="train", room_dimension_mode="invalid")
def test_unknown_height_stays_unknown_and_labels_are_not_input():
sample, source = pair(height=None)
result = project_sample(sample, source, split="train")
assert result["provenance"]["minimal_request"]["room_size_m"] == [4., 3.]
assert result["condition"]["room"]["height_m"] is None
assert "fixed_objects" not in result["condition"]["room"]
assert result["condition"]["constraints"] == []
assert set(result["condition"]["objects"][0]) == {"id", "category", "description"}
assert "target_size_local_m" not in json.dumps(result["condition"])
xyz = project_sample(sample, source, split="train", room_dimension_mode="xyz")
assert xyz["provenance"]["minimal_request"]["room_size_m"] == [4., 3.]
assert xyz["condition"]["room"]["height_m"] is None
def test_geometric_yaw_does_not_certify_semantic_front_or_snap_position():
sample, source = pair(floor=0.)
sample["target"]["objects"][0]["bottom_center_m"][2] = -.0005
source["objects"][0]["pos"][2] = -.0005
result = project_sample(sample, source, split="train")
assert result["validity"]["yaw"] == [True]
assert result["provenance"]["yaw_label_semantics"] == "local_bbox_axes_not_certified_semantic_front"
assert result["provenance"]["parent_yaw_validity"] == [False]
assert result["target"]["objects"][0]["bottom_center_m"][2] == -.0005
assert collate_samples([result], TinyTokenizer())["yaw_symmetry_order"].tolist() == [[2]]
def test_object_budget_rejects_whole_scene_without_partial_projection():
sample, source = pair()
before = deepcopy(sample)
with pytest.raises(ValueError, match="max_objects"):
project_sample(sample, source, split="train", max_objects=0)
assert sample == before
second = {**deepcopy(sample["condition"]["objects"][0]), "id": "obj_0001"}
sample["condition"]["objects"] = sample["condition"]["objects"] + [second]
with pytest.raises(ValueError, match="max_objects"):
project_sample(sample, source, split="train", max_objects=1)
def test_source_split_and_nonfinite_yaw_rejected():
sample, source = pair()
with pytest.raises(ValueError, match="split"):
project_sample(sample, source, split="test")
sample["target"]["objects"][0]["yaw_rad"] = float("nan")
with pytest.raises(ValueError, match="labels"):
project_sample(sample, source, split="train")
@pytest.mark.parametrize("field,value,reason", [
("room_type", "kitchen", "source_room_type"),
("height", 4., "source_room_height"),
("floor_z_m", 2., "source_floor_frame"),
("boundary_type", "hull", "source_boundary_semantics"),
])
def test_source_room_metadata_must_agree_with_parent(field, value, reason):
sample, source = pair()
source[field] = value
with pytest.raises(ValueError, match=reason):
project_sample(sample, source, split="train")
def test_parent_cannot_invent_height_or_hide_source_height_without_evidence():
sample, source = pair(height=None)
sample["condition"]["room"]["height_m"] = 3.
with pytest.raises(ValueError, match="source_room_height"):
project_sample(sample, source, split="train")
sample, source = pair(height=3.)
sample["condition"]["room"]["height_m"] = None
with pytest.raises(ValueError, match="source_room_height"):
project_sample(sample, source, split="train")
sample["provenance"]["legacy_height_dropped"] = True
result = project_sample(sample, source, split="train")
assert result["condition"]["room"]["height_m"] is None
assert result["provenance"]["qualification_source_height_m"] == 3.
def test_recorded_parent_height_drop_cannot_bypass_source_ceiling_qualification():
sample, source = pair(floor=0., height=2.)
sample["condition"]["room"]["height_m"] = None
sample["provenance"]["legacy_height_dropped"] = True
sample["target"]["objects"][0]["bottom_center_m"][2] = 1.5
source["objects"][0]["pos"][2] = 1.5
with pytest.raises(ValueError, match="ceiling"):
project_sample(sample, source, split="train")
def test_spatiallm_source_has_declared_canonical_zero_floor_by_default():
sample, source = pair(floor=0.)
del source["floor_z_m"]
assert project_sample(sample, source, split="train")["condition"]["room"]["floor_z_m"] == 0.
sample["condition"]["room"]["floor_z_m"] = .1
with pytest.raises(ValueError, match="source_floor_frame"):
project_sample(sample, source, split="train")
@pytest.mark.parametrize("change,reason", [
(lambda s, ir: s["provenance"].update(source="HSSD200"), "source"),
(lambda s, ir: s["condition"]["room"].update(boundary_known=False), "boundary"),
(lambda s, ir: s["condition"]["room"].update(floor_known=False), "floor"),
(lambda s, ir: s["condition"]["room"].update(room_type="unknown"), "room_type"),
(lambda s, ir: s["condition"]["room"].update(floor_polygon_xy_m=[[10., 20.], [14., 20.], [10., 23.]]), "rectangle"),
(lambda s, ir: s["validity"]["size"][0].__setitem__(2, False), "labels"),
(lambda s, ir: s["provenance"]["field_evidence"][0].update(tilted=True), "tilted"),
(lambda s, ir: ir["objects"][0].update(tilted=True), "tilted"),
(lambda s, ir: ir["objects"][0].update(yaw=.5), "source_geometry"),
(lambda s, ir: ir.update(uid="other"), "source_identity"),
])
def test_ineligible_geometry_fails_closed(change, reason):
sample, source = pair()
change(sample, source)
with pytest.raises(ValueError, match=reason):
project_sample(sample, source, split="train")
def test_obb_corners_and_own_height_determine_qualification():
sample, source = pair()
sample["target"]["objects"][0]["bottom_center_m"][0] = 10.25
source["objects"][0]["pos"][0] = 10.25
with pytest.raises(ValueError, match="xy_boundary"):
project_sample(sample, source, split="train")
sample, source = pair(height=1.)
with pytest.raises(ValueError, match="ceiling"):
project_sample(sample, source, split="train")
sample, source = pair()
with pytest.raises(ValueError, match="max_objects"):
project_sample(sample, source, split="train", max_objects=0)
def parent_files(tmp_path, duplicate_group=False):
parent, source_path = tmp_path / "parent", tmp_path / "source.jsonl"
parent.mkdir()
sources = []
for split in ("train", "validation", "test"):
row, source = pair(uid=split, split=split)
if duplicate_group:
row["provenance"]["house_id"] = row["provenance"]["group"] = "same_house"
source["group"] = "same_house"
(parent / (split + ".jsonl")).write_text(json.dumps(row) + "\n")
sources.append(source)
(parent / "manifest.json").write_text(json.dumps({"schema_version": "fastfill.v2"}))
source_path.write_text("".join(json.dumps(r) + "\n" for r in sources))
return parent, source_path
def test_build_inherits_split_and_does_not_change_parent_bytes(tmp_path):
parent, source = parent_files(tmp_path)
before = {p.name: p.read_bytes() for p in parent.iterdir()}
out = tmp_path / "new-version"
manifest = build_minimal_dataset(parent, source, out, expected_source_sha256=fingerprint(source))
assert manifest["split_samples"] == {"train": 1, "validation": 1, "test": 1}
assert manifest["split_objects"] == {"train": 1, "validation": 1, "test": 1}
assert manifest["room_dimension_mode"] == "xy"
assert manifest["split_integrity"]["group_cross_split_conflicts"] == 0
for split in ("train", "validation", "test"):
row = json.loads((out / (split + ".jsonl")).read_text())
assert row["provenance"]["scene_id"] == row["provenance"]["split"] == split
assert before == {p.name: p.read_bytes() for p in parent.iterdir()}
with pytest.raises(FileExistsError):
build_minimal_dataset(parent, source, out, expected_source_sha256=fingerprint(source))
def test_build_rejects_source_hash_and_cross_split_group(tmp_path):
parent, source = parent_files(tmp_path, duplicate_group=True)
with pytest.raises(ValueError, match="SHA256"):
build_minimal_dataset(parent, source, tmp_path / "bad-hash", expected_source_sha256="0" * 64)
assert not (tmp_path / "bad-hash").exists()
with pytest.raises(ValueError, match="cross.split"):
build_minimal_dataset(parent, source, tmp_path / "bad-split", expected_source_sha256=fingerprint(source))
|