File size: 13,922 Bytes
3fd7410 | 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 | """Layered IoU search using one or two pose-guided angle choices per axis."""
from __future__ import annotations
import argparse
import hashlib
import json
import platform
import sys
import time
from pathlib import Path
from typing import Any
from benchmark_core import (
Pose,
_euler_to_matrix,
_mat_mul,
get_last_render_backend,
iou,
render_views_fast,
)
from bfs_iou_oracle import (
ACTION_DIRECTIONS,
INVERSE_AXIS,
_matrix_to_euler_local,
matrix_key,
resolve_size,
score_batch_cuda,
)
try:
import torch
except Exception:
torch = None
ALLOWED_ANGLES = (30, 10, 5, 1)
CANONICAL_TASK_COUNT = 676
CANONICAL_TASKS_SHA256 = "c677e3cfdce7ddf3b6e2e9e932146e94e0ce66c52cd70940ebe0d310207896d9"
DEFAULT_OUTPUT = "artifacts/guided_iou_nearest1_676_v1/results.jsonl"
def sha256_file(path: Path) -> str:
digest = hashlib.sha256()
with path.open("rb") as handle:
for chunk in iter(lambda: handle.read(1024 * 1024), b""):
digest.update(chunk)
return digest.hexdigest()
def transpose(matrix):
return tuple(tuple(matrix[column][row] for column in range(3)) for row in range(3))
def guided_actions(matrix, target_matrix, nearest_count: int):
relative = _mat_mul(target_matrix, transpose(matrix))
residual_x, residual_y, residual_z = _matrix_to_euler_local(relative)
residuals = {
"PITCH": abs(residual_x),
"YAW": abs(residual_y),
"ROLL": abs(residual_z),
}
actions = []
for axis, rotation_fn in ACTION_DIRECTIONS:
axis_name = axis.split("_", 1)[0]
angles = sorted(ALLOWED_ANGLES, key=lambda value: (abs(value - residuals[axis_name]), -value))[:nearest_count]
for angle in angles:
actions.append((axis, angle, rotation_fn(angle)))
return actions
def reconstruct_path(parents, key):
path = []
while parents[key] is not None:
parent_key, axis, angle = parents[key]
path.append({"action": axis, "angle": angle})
key = parent_key
path.reverse()
return path
def search_task(
task: dict[str, Any],
*,
nearest_count: int,
max_depth: int,
max_states: int,
score_chunk_size: int,
) -> dict[str, Any]:
started = time.perf_counter()
target_pose = Pose.from_dict(task["target_pose"])
initial_pose = Pose.from_dict(task["init_pose"])
display_size = resolve_size(task.get("display_size"))
eval_size = int(task.get("eval_size", 320))
threshold = float(task.get("tau_success", 0.9))
_, target_eval, _ = render_views_fast(
task, target_pose, display_size=display_size, eval_size=eval_size, need_oblique=False
)
target_matrix = _euler_to_matrix(target_pose.roll, target_pose.pitch, target_pose.yaw)
initial_matrix = _euler_to_matrix(initial_pose.roll, initial_pose.pitch, initial_pose.yaw)
initial_key = matrix_key(initial_matrix)
frontier = [(initial_matrix, initial_key, None)]
visited = {initial_key}
parents = {initial_key: None}
layer_stats = []
best_iou = -1.0
best_depth = 0
if torch is not None and torch.cuda.is_available():
torch.cuda.reset_peak_memory_stats()
for depth in range(max_depth + 1):
layer_started = time.perf_counter()
success_index = None
success_iou = None
layer_best_iou = -1.0
scored_states = 0
for chunk_start in range(0, len(frontier), score_chunk_size):
chunk = frontier[chunk_start : chunk_start + score_chunk_size]
matrices = [item[0] for item in chunk]
scores = score_batch_cuda(task, matrices, target_eval)
if scores is None:
scores = []
for matrix in matrices:
roll, pitch, yaw = _matrix_to_euler_local(matrix)
_, current_eval, _ = render_views_fast(
task, Pose(roll, pitch, yaw), display_size=display_size, eval_size=eval_size, need_oblique=False
)
scores.append(float(iou(target_eval, current_eval)))
scored_states += len(scores)
layer_best_iou = max(layer_best_iou, max(scores))
for chunk_index, value in enumerate(scores):
if value >= threshold:
success_index = chunk_start + chunk_index
success_iou = value
break
if success_index is not None:
break
score_seconds = time.perf_counter() - layer_started
if layer_best_iou > best_iou:
best_iou = layer_best_iou
best_depth = depth
layer_stats.append({
"depth": depth,
"frontier_states": len(frontier),
"scored_states": scored_states,
"score_seconds": round(score_seconds, 6),
"best_iou": round(layer_best_iou, 9),
})
if success_index is not None:
success_key = frontier[success_index][1]
elapsed = time.perf_counter() - started
return {
"sample_id": task["sample_id"],
"nearest_angle_count": nearest_count,
"search_status": "success",
"optimal_steps_under_guided_actions": depth,
"success_iou": round(float(success_iou), 9),
"path": reconstruct_path(parents, success_key),
"expanded_states": sum(row["scored_states"] for row in layer_stats),
"visited_states": len(visited),
"elapsed_seconds": round(elapsed, 6),
"seconds_per_search_depth": round(elapsed / max(1, depth), 6),
"layer_stats": layer_stats,
"gpu_peak_reserved_gb": round(torch.cuda.max_memory_reserved() / 1024**3, 6)
if torch is not None and torch.cuda.is_available() else 0.0,
}
if depth >= max_depth:
break
generation_started = time.perf_counter()
next_frontier = []
for matrix, key, previous_action in frontier:
for axis, angle, delta in guided_actions(matrix, target_matrix, nearest_count):
if previous_action == (INVERSE_AXIS[axis], angle):
continue
next_matrix = _mat_mul(delta, matrix)
next_key = matrix_key(next_matrix)
if next_key in visited:
continue
visited.add(next_key)
parents[next_key] = (key, axis, angle)
next_frontier.append((next_matrix, next_key, (axis, angle)))
if len(visited) >= max_states:
elapsed = time.perf_counter() - started
return {
"sample_id": task["sample_id"],
"nearest_angle_count": nearest_count,
"search_status": "state_limit_reached",
"optimal_steps_under_guided_actions": None,
"best_iou": round(best_iou, 9),
"best_depth": best_depth,
"expanded_states": sum(row["scored_states"] for row in layer_stats),
"visited_states": len(visited),
"elapsed_seconds": round(elapsed, 6),
"layer_stats": layer_stats,
}
layer_stats[-1]["generation_seconds"] = round(time.perf_counter() - generation_started, 6)
layer_stats[-1]["next_frontier_states"] = len(next_frontier)
frontier = next_frontier
if not frontier:
break
elapsed = time.perf_counter() - started
return {
"sample_id": task["sample_id"],
"nearest_angle_count": nearest_count,
"search_status": "not_found_within_depth",
"optimal_steps_under_guided_actions": None,
"best_iou": round(best_iou, 9),
"best_depth": best_depth,
"expanded_states": sum(row["scored_states"] for row in layer_stats),
"visited_states": len(visited),
"elapsed_seconds": round(elapsed, 6),
"layer_stats": layer_stats,
}
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--tasks-file", default="data/tasks_674.jsonl")
parser.add_argument("--ids", default="", help="Comma-separated subset; empty means all canonical tasks")
parser.add_argument("--nearest-count", type=int, choices=(1, 2), default=1)
parser.add_argument("--max-depth", type=int, default=12)
parser.add_argument("--max-states", type=int, default=2_000_000)
parser.add_argument("--candidate-batch-size", type=int, default=32)
parser.add_argument("--point-batch-size", type=int, default=100_000)
parser.add_argument("--score-chunk-size", type=int, default=4096)
parser.add_argument("--gpu-memory-fraction", type=float, default=0.48)
parser.add_argument("--output", default=DEFAULT_OUTPUT)
args = parser.parse_args()
if torch is not None and torch.cuda.is_available():
torch.cuda.set_per_process_memory_fraction(args.gpu_memory_fraction, 0)
tasks_path = Path(args.tasks_file)
tasks_sha256 = sha256_file(tasks_path)
if tasks_sha256 != CANONICAL_TASKS_SHA256:
raise RuntimeError("Canonical tasks SHA-256 mismatch")
selected_ids = {value.strip() for value in args.ids.split(",") if value.strip()}
tasks = []
source_ids = []
with tasks_path.open(encoding="utf-8") as handle:
for line in handle:
task = json.loads(line)
source_ids.append(task["sample_id"])
if not selected_ids or task["sample_id"] in selected_ids:
task["bfs_candidate_batch_size"] = args.candidate_batch_size
task["cuda_point_batch_size"] = args.point_batch_size
tasks.append(task)
if len(source_ids) != CANONICAL_TASK_COUNT or len(set(source_ids)) != CANONICAL_TASK_COUNT:
raise RuntimeError("Canonical source must contain exactly 676 unique tasks")
tasks.sort(key=lambda task: task["sample_id"])
if selected_ids and {task["sample_id"] for task in tasks} != selected_ids:
raise RuntimeError("Requested task IDs were not found")
output = Path(args.output)
output.parent.mkdir(parents=True, exist_ok=True)
manifest = {
"protocol_version": "guided_iou_nearest1_v1" if args.nearest_count == 1 else "guided_iou_nearest2_comparison_v1",
"tasks_file": str(tasks_path).replace("\\", "/"),
"tasks_sha256": tasks_sha256,
"canonical_task_count": CANONICAL_TASK_COUNT,
"selected_task_count": len(tasks),
"selected_ids": [task["sample_id"] for task in tasks] if selected_ids else "all_canonical_tasks",
"search": {
"directions": [axis for axis, _ in ACTION_DIRECTIONS],
"allowed_angles_degrees": list(ALLOWED_ANGLES),
"nearest_angle_count_per_axis": args.nearest_count,
"relative_pose_method": "rotation_matrix",
"success_condition": "IoU >= task tau_success (canonical value 0.9)",
"algorithm": "strict layer-by-layer BFS; first successful layer is optimal under guided actions",
"max_depth": args.max_depth,
"max_states": args.max_states,
},
"runtime": {
"platform": platform.platform(),
"python": sys.version,
"torch": getattr(torch, "__version__", None),
"torch_cuda": getattr(getattr(torch, "version", None), "cuda", None),
"cuda_available": bool(torch is not None and torch.cuda.is_available()),
"gpu_name": torch.cuda.get_device_name(0) if torch is not None and torch.cuda.is_available() else None,
"gpu_memory_fraction": args.gpu_memory_fraction,
"candidate_batch_size": args.candidate_batch_size,
"point_batch_size": args.point_batch_size,
"score_chunk_size": args.score_chunk_size,
},
"scope_note": "Pose-guided constrained shortest path, not the global optimum over the full 24-action space.",
}
manifest_path = output.with_suffix(".manifest.json")
if manifest_path.exists():
existing_manifest = json.loads(manifest_path.read_text(encoding="utf-8"))
if existing_manifest != manifest:
raise RuntimeError(f"Run manifest mismatch: {manifest_path}")
else:
manifest_path.write_text(json.dumps(manifest, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
completed = set()
if output.exists():
existing_rows = [json.loads(line) for line in output.read_text(encoding="utf-8").splitlines() if line.strip()]
if any(row.get("nearest_angle_count") != args.nearest_count for row in existing_rows):
raise RuntimeError("Existing results use a different nearest-angle count")
completed = {row["sample_id"] for row in existing_rows}
with output.open("a", encoding="utf-8") as handle:
for task in tasks:
if task["sample_id"] in completed:
continue
print(json.dumps({"event": "start", "sample_id": task["sample_id"], "nearest_count": args.nearest_count}), flush=True)
result = search_task(
task,
nearest_count=args.nearest_count,
max_depth=args.max_depth,
max_states=args.max_states,
score_chunk_size=args.score_chunk_size,
)
result["render_backend"] = get_last_render_backend()
result["scoring_backend"] = (
"cuda_batch" if torch is not None and torch.cuda.is_available() else "renderer_per_state"
)
handle.write(json.dumps(result, ensure_ascii=False) + "\n")
handle.flush()
print(json.dumps(result, ensure_ascii=False), flush=True)
if __name__ == "__main__":
main()
|