File size: 14,219 Bytes
116e377 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 | """Build twelve NeoVerse-style camera controls for every SpatialVID scene.
The relative trajectories reproduce ``CameraTrajectory.from_predefined`` in
NeoVerse's ``diffsynth/utils/auxiliary.py``. SpatialVID is a dynamic source,
so the NeoVerse demo composition is applied frame by frame:
target_c2w[t] = source_c2w[t] @ relative_c2w[t]
The emitted controls use the unified dataset's absolute OpenCV c2w schema.
"""
from __future__ import annotations
import argparse
import json
from copy import deepcopy
from pathlib import Path
from typing import Any
import numpy as np
from scipy.spatial.transform import Rotation, Slerp
ROOT = Path(__file__).resolve().parent
NUM_FRAMES = 81
LAST_FRAME = NUM_FRAMES - 1
CONTROL_SPECS: dict[str, dict[str, float]] = {
"pan_left": {"angle": 15.0},
"pan_right": {"angle": 15.0},
"tilt_up": {"angle": 15.0},
"tilt_down": {"angle": 15.0},
"move_left": {"distance": 0.1},
"move_right": {"distance": 0.1},
"push_in": {"distance": 0.1},
"pull_out": {"distance": 0.1},
"boom_up": {"distance": 0.1},
"boom_down": {"distance": 0.1},
"orbit_left": {"angle": 15.0, "orbit_radius": 1.0},
"orbit_right": {"angle": 15.0, "orbit_radius": 1.0},
}
COMPOSITION = "target_c2w[t] = source_c2w[t] @ relative_c2w[t]"
PROVENANCE = (
"NeoVerse CameraTrajectory.from_predefined defaults and dynamic-scene "
"relative composition"
)
def relative_endpoint(name: str, params: dict[str, float]) -> np.ndarray:
"""Match NeoVerse CameraTrajectory._get_camera_matrix exactly."""
angle = np.deg2rad(params.get("angle", 30.0))
radius = params.get("orbit_radius", 0.5)
pose = np.eye(4, dtype=np.float64)
if name == "move_left":
pose[0, 3] = -params["distance"]
elif name == "move_right":
pose[0, 3] = params["distance"]
elif name == "push_in":
pose[2, 3] = params["distance"]
elif name == "pull_out":
pose[2, 3] = -params["distance"]
elif name == "boom_up":
pose[1, 3] = -params["distance"]
elif name == "boom_down":
pose[1, 3] = params["distance"]
elif name in {"orbit_left", "orbit_right"}:
theta = -angle if name == "orbit_left" else angle
cosine, sine = np.cos(theta), np.sin(theta)
pose[0, 3] = radius * sine
pose[2, 3] = radius * (1.0 - cosine)
pose[0, 0] = cosine
pose[0, 2] = -sine
pose[2, 0] = sine
pose[2, 2] = cosine
elif name in {"pan_left", "pan_right"}:
theta = -angle if name == "pan_left" else angle
cosine, sine = np.cos(theta), np.sin(theta)
pose[0, 0] = cosine
pose[0, 2] = sine
pose[2, 0] = -sine
pose[2, 2] = cosine
elif name in {"tilt_up", "tilt_down"}:
theta = angle if name == "tilt_up" else -angle
cosine, sine = np.cos(theta), np.sin(theta)
pose[1, 1] = cosine
pose[1, 2] = -sine
pose[2, 1] = sine
pose[2, 2] = cosine
else:
raise ValueError(f"unsupported control: {name}")
return pose
def relative_trajectory(name: str, params: dict[str, float]) -> np.ndarray:
"""Match NeoVerse's two-keyframe rotation Slerp and linear translation."""
endpoint = relative_endpoint(name, params)
frame_times = np.linspace(0.0, 1.0, NUM_FRAMES)
rotations = Slerp(
[0.0, 1.0],
Rotation.from_matrix(np.stack([np.eye(3), endpoint[:3, :3]])),
)(frame_times).as_matrix()
positions = frame_times[:, None] * endpoint[:3, 3][None]
output = np.repeat(np.eye(4, dtype=np.float64)[None], NUM_FRAMES, axis=0)
output[:, :3, :3] = rotations
output[:, :3, 3] = positions
return output
def trajectory_spec(name: str, params: dict[str, float]) -> dict[str, Any]:
json_params = {
key: int(value) if key == "angle" and float(value).is_integer() else value
for key, value in params.items()
}
return {
"name": name,
"description": f"NeoVerse predefined {name} control for SpatialVID",
"mode": "relative",
"num_frames": NUM_FRAMES,
"zoom_ratio": 1.0,
"use_first_frame": True,
"composition": COMPOSITION,
"keyframes": [
{"0": [{"static": {}}]},
{str(LAST_FRAME): [{name: json_params}]},
],
}
def source_records(manifest: dict[str, Any]) -> list[dict[str, Any]]:
"""Recover one authoritative source record per scene, also after reruns."""
records: dict[str, dict[str, Any]] = {}
for case in manifest["cases"]:
scene = case["scene"]
if scene in records:
continue
base = deepcopy(case)
base["case_id"] = scene
base["traj"] = None
base["target_path"] = None
base["camera_control"] = None
base["pair_id"] = None
metadata = deepcopy(base.get("metadata", {}))
metadata.pop("camera_control_provenance", None)
base["metadata"] = metadata
records[scene] = base
return [records[key] for key in sorted(records)]
def control_document(source: dict[str, Any], relative: np.ndarray) -> dict[str, Any]:
frames = source["frames"]
if len(frames) != NUM_FRAMES:
raise ValueError(f"expected {NUM_FRAMES} source frames, got {len(frames)}")
source_c2w = np.asarray([frame["c2w"] for frame in frames], dtype=np.float64)
target_c2w = source_c2w @ relative
return {
"format_version": source["format_version"],
"camera_model": source["camera_model"],
"pose_type": source["pose_type"],
"coordinate_system": source["coordinate_system"],
"image_hw": source["image_hw"],
"num_frames": NUM_FRAMES,
"frames": [
{
"frame_index": index,
"c2w": target_c2w[index].tolist(),
"intrinsics": deepcopy(frames[index]["intrinsics"]),
}
for index in range(NUM_FRAMES)
],
}
def controlled_case(base: dict[str, Any], name: str, params: dict[str, float]) -> dict[str, Any]:
scene = base["scene"]
case = deepcopy(base)
case["case_id"] = f"{scene}__{name}"
case["traj"] = name
case["target_path"] = None
case["camera_control"] = {
"path": f"camera_controls/{scene}/{name}/transforms.json",
"variant": "default",
}
case["pair_id"] = None
case["metadata"]["camera_control_provenance"] = {
"source": PROVENANCE,
"spec_path": f"trajectory_specs/{name}.json",
"mode": "relative",
"composition": COMPOSITION,
"angle_degrees": params.get("angle"),
"distance": params.get("distance"),
"distance_unit": "source_reconstruction_unit" if "distance" in params else None,
"orbit_radius": params.get("orbit_radius"),
"orbit_radius_unit": "source_reconstruction_unit" if "orbit_radius" in params else None,
}
return case
def update_manifest(manifest: dict[str, Any], bases: list[dict[str, Any]]) -> dict[str, Any]:
output = deepcopy(manifest)
cases = [
controlled_case(base, name, params)
for base in bases
for name, params in CONTROL_SPECS.items()
]
output["protocol"]["control"] = "camera"
output["camera"].update({
"default_control_variant": "default",
"synthetic_control_mode": "relative",
"synthetic_control_composition": COMPOSITION,
})
output["extensions"]["camera_trajectory_specs"] = {
"path": "trajectory_specs/<trajectory_id>.json",
"meaning": "NeoVerse-compatible relative keyframe trajectory definition",
"source": PROVENANCE,
}
output["statistics"].update({
"num_scenes": len(bases),
"num_cases": len(cases),
"num_camera_control_cases": len(cases),
"num_camera_control_frames": len(cases) * NUM_FRAMES,
"camera_control_trajectory_distribution": {
name: len(bases) for name in CONTROL_SPECS
},
"num_pairs": 0,
"num_unpaired_cases": len(cases),
})
output["pairing"]["unpaired_sources"] = [
{
"case_id": case["case_id"],
"source_scene_id": case["scene"],
"source_path": case["source_path"],
"num_frames": NUM_FRAMES,
"reason": "synthetic_camera_control_only_no_target_rgb",
}
for case in cases
]
output["cases"] = cases
return output
def validate(root: Path) -> dict[str, Any]:
manifest = json.loads((root / "manifest.json").read_text(encoding="utf-8"))
errors: list[str] = []
max_first_error = 0.0
max_composition_error = 0.0
max_orthonormal_error = 0.0
max_determinant_error = 0.0
relative = {
name: relative_trajectory(name, params)
for name, params in CONTROL_SPECS.items()
}
case_ids = [case["case_id"] for case in manifest["cases"]]
control_paths = [case["camera_control"]["path"] for case in manifest["cases"]]
if len(case_ids) != len(set(case_ids)):
errors.append("duplicate case_id")
if len(control_paths) != len(set(control_paths)):
errors.append("duplicate camera control path")
for case in manifest["cases"]:
control_path = root / case["camera_control"]["path"]
source_path = root / case["source_path"] / "transforms.json"
if not control_path.is_file():
errors.append(f"missing {control_path.relative_to(root)}")
continue
control = json.loads(control_path.read_text(encoding="utf-8"))
source = json.loads(source_path.read_text(encoding="utf-8"))
if len(control["frames"]) != NUM_FRAMES:
errors.append(f"{case['case_id']}: wrong frame count")
continue
control_c2w = np.asarray([frame["c2w"] for frame in control["frames"]])
source_c2w = np.asarray([frame["c2w"] for frame in source["frames"]])
max_first_error = max(max_first_error, float(np.max(np.abs(control_c2w[0] - source_c2w[0]))))
expected_c2w = source_c2w @ relative[case["traj"]]
max_composition_error = max(
max_composition_error,
float(np.max(np.abs(control_c2w - expected_c2w))),
)
rotations = control_c2w[:, :3, :3]
orth = rotations @ np.swapaxes(rotations, 1, 2)
max_orthonormal_error = max(
max_orthonormal_error,
float(np.max(np.abs(orth - np.eye(3)[None]))),
)
max_determinant_error = max(
max_determinant_error,
float(np.max(np.abs(np.linalg.det(rotations) - 1.0))),
)
if not np.isfinite(control_c2w).all():
errors.append(f"{case['case_id']}: non-finite pose")
if any(frame["intrinsics"]["fx"] <= 0 or frame["intrinsics"]["fy"] <= 0 for frame in control["frames"]):
errors.append(f"{case['case_id']}: non-positive focal length")
if [frame["frame_index"] for frame in control["frames"]] != list(range(NUM_FRAMES)):
errors.append(f"{case['case_id']}: invalid frame indexes")
if [frame["intrinsics"] for frame in control["frames"]] != [
frame["intrinsics"] for frame in source["frames"]
]:
errors.append(f"{case['case_id']}: intrinsics differ from source")
if len(manifest["cases"]) != len(CONTROL_SPECS) * manifest["statistics"]["num_scenes"]:
errors.append("manifest case count does not match scene × trajectory count")
if max_first_error > 1e-9:
errors.append(f"first-frame mismatch: {max_first_error}")
if max_composition_error > 1e-9:
errors.append(f"composition mismatch: {max_composition_error}")
if max_orthonormal_error > 1e-6:
errors.append(f"rotation orthonormal error: {max_orthonormal_error}")
if max_determinant_error > 1e-6:
errors.append(f"rotation determinant error: {max_determinant_error}")
if errors:
raise RuntimeError("\n".join(errors[:20]))
return {
"scenes": manifest["statistics"]["num_scenes"],
"cases": len(manifest["cases"]),
"trajectories": list(CONTROL_SPECS),
"max_first_frame_pose_error": max_first_error,
"max_composition_error": max_composition_error,
"max_rotation_orthonormal_error": max_orthonormal_error,
"max_rotation_determinant_error": max_determinant_error,
}
def build(root: Path) -> dict[str, Any]:
manifest_path = root / "manifest.json"
manifest = json.loads(manifest_path.read_text(encoding="utf-8"))
bases = source_records(manifest)
if len(bases) != 200:
raise ValueError(f"expected 200 SpatialVID scenes, found {len(bases)}")
spec_root = root / "trajectory_specs"
spec_root.mkdir(parents=True, exist_ok=True)
relative = {}
for name, params in CONTROL_SPECS.items():
(spec_root / f"{name}.json").write_text(
json.dumps(trajectory_spec(name, params), indent=2) + "\n",
encoding="utf-8",
)
relative[name] = relative_trajectory(name, params)
for base in bases:
scene = base["scene"]
source_path = root / base["source_path"] / "transforms.json"
source = json.loads(source_path.read_text(encoding="utf-8"))
for name in CONTROL_SPECS:
output_path = root / "camera_controls" / scene / name / "transforms.json"
output_path.parent.mkdir(parents=True, exist_ok=True)
output_path.write_text(
json.dumps(control_document(source, relative[name]), indent=2) + "\n",
encoding="utf-8",
)
manifest_temp = manifest_path.with_suffix(".json.tmp")
manifest_temp.write_text(
json.dumps(update_manifest(manifest, bases), indent=2, ensure_ascii=False) + "\n",
encoding="utf-8",
)
manifest_temp.replace(manifest_path)
return validate(root)
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--root", type=Path, default=ROOT)
parser.add_argument("--check", action="store_true")
args = parser.parse_args()
result = validate(args.root) if args.check else build(args.root)
print(json.dumps(result, indent=2))
if __name__ == "__main__":
main()
|