File size: 25,046 Bytes
540e67a 3c21768 540e67a a8124a8 540e67a a8124a8 540e67a 3c21768 540e67a 3c21768 540e67a eeac43c 3c21768 eeac43c 540e67a 3c21768 540e67a 3c21768 540e67a 3c21768 540e67a eeac43c 3c21768 eeac43c 540e67a 3c21768 540e67a 3c21768 540e67a 3c21768 540e67a 45c1706 540e67a 3c21768 540e67a eeac43c 3c21768 eeac43c 540e67a 3c21768 540e67a 3c21768 540e67a eeac43c 3c21768 eeac43c 540e67a 3c21768 540e67a eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a a8124a8 540e67a b7a466b eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a a8124a8 540e67a eeac43c 540e67a a8124a8 540e67a 3c21768 540e67a | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 | {
"source": "docs/data/task_suite_20.json plus results/episode_task_suite/summary_report.json",
"dataset_scope": {
"sample_episode_count": 1,
"num_frames": 5821,
"num_windows": 1161,
"feature_dim": 8546,
"warning": "Single public sample episode; this supports pipeline/task evidence, while cross-episode generalization requires held-out episodes."
},
"baselines": {
"minimal": "Interpretable softmax, logistic, ridge, and retrieval heads over the 8,546-d window feature vector.",
"neural_mlp": "Small PyTorch MLP classifiers/regressors using the same features, splits, and task contracts."
},
"task_count": 20,
"directions": {
"A": {
"id": "human_motion",
"name": "Human Modeling & Motion Understanding",
"focus": "Human/hand/body motion, deformation priors, human-object interaction, affordance modeling.",
"preferred_background": "Human pose/shape estimation, SMPL-style models, motion capture, or motion generation.",
"current_status": "partially implemented",
"current_readout": "The sample supports hand trajectory forecasting and contact/object probes, but it does not yet include a full body/shape model or multi-person priors.",
"next_steps": [
"Add SMPL/SMPL-X or MANO-style body/hand parameter targets where available.",
"Train sequence models over multi-episode motion trajectories instead of isolated windows.",
"Evaluate affordance prediction on held-out objects and held-out episodes."
],
"tasks": [
"timeline_action",
"hand_trajectory_forecast",
"contact_prediction",
"object_relevance",
"interaction_text_prediction",
"imu_to_hand_pose"
],
"task_display_names": [
"Action Recognition",
"Hand Trajectory Forecasting",
"Contact State Prediction",
"Object Relevance Prediction",
"Interaction Text Prediction",
"IMU-to-Hand Pose Reconstruction"
],
"counts": {
"direct": 3,
"proxy": 3,
"diagnostic": 0,
"total_links": 6
}
},
"B": {
"id": "reconstruction_rendering",
"name": "3D/4D Reconstruction & Neural Rendering",
"focus": "Multi-view dynamic scene reconstruction, NeRF/Gaussian Splatting, novel-view synthesis.",
"preferred_background": "3D reconstruction, neural rendering, camera calibration, and bundle adjustment.",
"current_status": "proxy tasks only",
"current_readout": "The current suite checks cross-modal alignment and depth/video reconstruction proxies; it does not yet train a renderer or reconstruct geometry.",
"next_steps": [
"Use calibrated multi-view video plus SLAM pose to build per-episode camera trajectories.",
"Add depth-supervised point clouds, TSDF, Gaussian Splatting, or NeRF baselines.",
"Evaluate novel-view synthesis and temporal consistency across held-out views/time."
],
"tasks": [
"cross_modal_retrieval",
"modality_reconstruction",
"misalignment_detection",
"imu_to_hand_pose",
"camera_view_sync_retrieval"
],
"task_display_names": [
"Cross-Modal Retrieval",
"Cross-Modal Reconstruction",
"Multimodal Synchronization Detection",
"IMU-to-Hand Pose Reconstruction",
"Camera-View Synchronization Retrieval"
],
"counts": {
"direct": 1,
"proxy": 3,
"diagnostic": 1,
"total_links": 5
}
},
"C": {
"id": "egocentric_interaction",
"name": "Egocentric Vision & Interaction",
"focus": "Egocentric action and intention understanding, hand-object interaction, gaze/attention modeling, task structure modeling.",
"preferred_background": "Video understanding, action recognition, or egocentric vision.",
"current_status": "strongest implemented track",
"current_readout": "The unified 20-task suite directly targets egocentric action, task state, interaction, grounding, forecasting, and alignment.",
"next_steps": [
"Move from single-episode chronological splits to held-out-episode splits.",
"Use audio together with stronger multimodal backbones for action, intent, and grounding.",
"Evaluate long-horizon task success prediction and action-conditioned generation."
],
"tasks": [
"timeline_action",
"timeline_subtask",
"transition_detection",
"next_action",
"hand_trajectory_forecast",
"contact_prediction",
"object_relevance",
"caption_grounding",
"cross_modal_retrieval",
"temporal_order",
"misalignment_detection",
"long_horizon_next_action",
"next_subtask_forecast",
"interaction_text_prediction",
"action_object_relation",
"object_set_forecast",
"time_to_transition"
],
"task_display_names": [
"Action Recognition",
"Procedure Step Recognition",
"Action Boundary Detection",
"Next-Action Prediction",
"Hand Trajectory Forecasting",
"Contact State Prediction",
"Object Relevance Prediction",
"Language Grounding",
"Cross-Modal Retrieval",
"Temporal Order Verification",
"Multimodal Synchronization Detection",
"Long-Horizon Next-Action Forecasting",
"Long-Horizon Next-Subtask Forecasting",
"Interaction Text Prediction",
"Action-Object Relation Prediction",
"Future Object-Set Forecasting",
"Time-to-Next-Transition Regression"
],
"counts": {
"direct": 10,
"proxy": 3,
"diagnostic": 4,
"total_links": 17
}
},
"D": {
"id": "world_modeling",
"name": "Scene Reconstruction & World Modeling",
"focus": "Long-term consistent 3D/4D scene mapping, scene graphs, object- and space-centric representations, spatial reasoning.",
"preferred_background": "Large-scale mapping, semantic reconstruction, or agent world models.",
"current_status": "early proxy tasks",
"current_readout": "The current tasks probe temporal structure, object relevance, cross-modal retrieval, and modality prediction, but they do not yet build persistent maps or scene graphs.",
"next_steps": [
"Convert windows into persistent object/scene-state nodes with timestamps and camera poses.",
"Add map consistency, object permanence, and spatial relation prediction tasks.",
"Train held-out-episode world models that predict future observations and task state."
],
"tasks": [
"timeline_subtask",
"transition_detection",
"next_action",
"object_relevance",
"caption_grounding",
"cross_modal_retrieval",
"modality_reconstruction",
"temporal_order",
"misalignment_detection",
"long_horizon_next_action",
"next_subtask_forecast",
"action_object_relation",
"object_set_forecast",
"camera_view_sync_retrieval",
"time_to_transition"
],
"task_display_names": [
"Procedure Step Recognition",
"Action Boundary Detection",
"Next-Action Prediction",
"Object Relevance Prediction",
"Language Grounding",
"Cross-Modal Retrieval",
"Cross-Modal Reconstruction",
"Temporal Order Verification",
"Multimodal Synchronization Detection",
"Long-Horizon Next-Action Forecasting",
"Long-Horizon Next-Subtask Forecasting",
"Action-Object Relation Prediction",
"Future Object-Set Forecasting",
"Camera-View Synchronization Retrieval",
"Time-to-Next-Transition Regression"
],
"counts": {
"direct": 1,
"proxy": 10,
"diagnostic": 4,
"total_links": 15
}
}
},
"tasks": {
"timeline_action": {
"name": "Timeline action recognition",
"family": "supervised",
"input": "all featurized modalities",
"output": "current action label",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"A": "proxy"
},
"why": "Reads egocentric sensor state as the current human action; also provides a weak human-motion readout.",
"current_limit": "Chronological single-episode split creates unseen future action classes.",
"display_name": "Action Recognition",
"artifact_id": "timeline_action",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 0.05,
"neural_mlp": 0.014814814814814814,
"better_baseline": "minimal"
}
},
"timeline_subtask": {
"name": "Timeline subtask recognition",
"family": "supervised",
"input": "all featurized modalities",
"output": "current subtask label",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"D": "proxy"
},
"why": "Segments egocentric task state and provides a first proxy for symbolic world/task state.",
"current_limit": "Single-episode ordering makes future subtasks hard to generalize.",
"display_name": "Procedure Step Recognition",
"artifact_id": "timeline_subtask",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 0.05056355513846935,
"neural_mlp": 0.02810810810810811,
"better_baseline": "minimal"
}
},
"transition_detection": {
"name": "Action transition detection",
"family": "diagnostic",
"input": "all featurized modalities",
"output": "boundary vs steady state",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"D": "diagnostic"
},
"why": "Localizes egocentric task boundaries and diagnoses temporal state changes.",
"current_limit": "Boundary class is sparse, so accuracy alone is misleading.",
"display_name": "Action Boundary Detection",
"artifact_id": "transition_detection",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 0.6118237590630229,
"neural_mlp": 0.5862068965517241,
"better_baseline": "minimal"
}
},
"next_action": {
"name": "Short-horizon next action",
"family": "supervised",
"input": "current multimodal window",
"output": "action 20 frames later",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"D": "proxy"
},
"why": "Tests action intention/task-flow prediction from egocentric context.",
"current_limit": "Unseen future labels dominate the single-episode chronological test.",
"display_name": "Next-Action Prediction",
"artifact_id": "next_action",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 0.05925925925925927,
"neural_mlp": 0.04186046511627907,
"better_baseline": "minimal"
}
},
"hand_trajectory_forecast": {
"name": "Hand trajectory forecasting",
"family": "forecast",
"input": "current multimodal window",
"output": "future left/right hand 3D joints",
"primary_direction": "A",
"direction_roles": {
"A": "direct",
"C": "proxy"
},
"why": "Directly predicts human hand motion and supports hand-object interaction modeling.",
"current_limit": "Forecasting is window-level and not yet a full sequence or policy model.",
"display_name": "Hand Trajectory Forecasting",
"artifact_id": "hand_trajectory_forecast",
"metric": {
"key": "mpjpe",
"name": "MPJPE",
"direction": "lower",
"minimal": 0.8646570444107056,
"neural_mlp": 0.10785018652677536,
"better_baseline": "neural_mlp"
}
},
"contact_prediction": {
"name": "Body/object contact prediction",
"family": "supervised",
"input": "non-contact/non-caption features",
"output": "binary contact label",
"primary_direction": "A",
"direction_roles": {
"A": "direct",
"C": "proxy"
},
"why": "Targets physical interaction state, a core affordance and manipulation signal.",
"current_limit": "The public sample is degenerate for this target because one class dominates.",
"display_name": "Contact State Prediction",
"artifact_id": "contact_prediction",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 1.0,
"neural_mlp": 1.0,
"better_baseline": "tie"
}
},
"object_relevance": {
"name": "Relevant object set prediction",
"family": "supervised",
"input": "non-caption feature blocks",
"output": "multi-label object set",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"A": "proxy",
"D": "proxy"
},
"why": "Connects egocentric activity to manipulated objects and early object-centric state.",
"current_limit": "Object labels are language-derived and sparse in one episode.",
"display_name": "Object Relevance Prediction",
"artifact_id": "object_relevance",
"metric": {
"key": "micro_f1",
"name": "micro-F1",
"direction": "higher",
"minimal": 0.18034382095361662,
"neural_mlp": 0.1679279279279279,
"better_baseline": "minimal"
}
},
"caption_grounding": {
"name": "Caption-to-window grounding",
"family": "retrieval",
"input": "caption objects/interaction query and candidate sensor windows",
"output": "matching time window",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"D": "proxy"
},
"why": "Grounds language annotation into egocentric sensor time and task state.",
"current_limit": "Bag-of-objects language features are too weak for rich grounding.",
"display_name": "Language Grounding",
"artifact_id": "caption_grounding",
"metric": {
"key": "mrr",
"name": "MRR",
"direction": "higher",
"minimal": 0.016023479050338015,
"neural_mlp": 0.01684125567132316,
"better_baseline": "neural_mlp"
}
},
"cross_modal_retrieval": {
"name": "Cross-modal retrieval",
"family": "retrieval",
"input": "motion/IMU/camera query",
"output": "matching depth/video window",
"primary_direction": "C",
"direction_roles": {
"C": "diagnostic",
"B": "proxy",
"D": "proxy"
},
"why": "Tests whether synchronized modalities identify the same 4D moment, a prerequisite for reconstruction and world modeling.",
"current_limit": "Retrieval shows an alignment signal, not geometric reconstruction.",
"display_name": "Cross-Modal Retrieval",
"artifact_id": "cross_modal_retrieval",
"metric": {
"key": "mrr",
"name": "MRR",
"direction": "higher",
"minimal": 0.26925966892956127,
"neural_mlp": 0.1299971898648288,
"better_baseline": "minimal"
}
},
"modality_reconstruction": {
"name": "Modality reconstruction",
"family": "forecast",
"input": "motion/IMU/camera",
"output": "depth/video feature vector",
"primary_direction": "B",
"direction_roles": {
"B": "proxy",
"D": "proxy"
},
"why": "Predicts visual/depth state from non-target sensors as a weak reconstruction/world-model objective.",
"current_limit": "Feature-vector reconstruction is not pixel, depth-map, mesh, NeRF, or Gaussian reconstruction.",
"display_name": "Cross-Modal Reconstruction",
"artifact_id": "modality_reconstruction",
"metric": {
"key": "r2",
"name": "R2",
"direction": "higher",
"minimal": -0.015271898913936655,
"neural_mlp": -0.010171410134180991,
"better_baseline": "neural_mlp"
}
},
"temporal_order": {
"name": "Temporal order verification",
"family": "diagnostic",
"input": "two adjacent windows",
"output": "correct vs reversed order",
"primary_direction": "C",
"direction_roles": {
"C": "diagnostic",
"D": "diagnostic"
},
"why": "Checks whether features encode local time direction and task progression.",
"current_limit": "Only local adjacent ordering, not long-horizon causal modeling.",
"display_name": "Temporal Order Verification",
"artifact_id": "temporal_order",
"metric": {
"key": "f1",
"name": "F1",
"direction": "higher",
"minimal": 0.5399515738498789,
"neural_mlp": 0.8520179372197308,
"better_baseline": "neural_mlp"
}
},
"misalignment_detection": {
"name": "Cross-modal misalignment detection",
"family": "diagnostic",
"input": "motion plus visual/depth pair",
"output": "aligned vs shifted",
"primary_direction": "C",
"direction_roles": {
"C": "diagnostic",
"B": "diagnostic",
"D": "diagnostic"
},
"why": "Detects temporal desynchronization, a key data-quality gate for multimodal reconstruction and world models.",
"current_limit": "Synthetic shifts diagnose alignment but do not solve calibration or mapping.",
"display_name": "Multimodal Synchronization Detection",
"artifact_id": "misalignment_detection",
"metric": {
"key": "f1",
"name": "F1",
"direction": "higher",
"minimal": 0.5051698670605613,
"neural_mlp": 0.7152682255845944,
"better_baseline": "neural_mlp"
}
},
"long_horizon_next_action": {
"name": "Long-horizon next-action forecasting",
"family": "classification",
"input": "current and historical windows",
"output": "future action label",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"D": "proxy"
},
"why": "Extends short-horizon intention prediction into longer activity futures, a key egocentric and world-model signal.",
"current_limit": "Evaluated from sample-supported future labels, not full open-world action generation.",
"display_name": "Long-Horizon Next-Action Forecasting",
"artifact_id": "long_horizon_next_action",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 0.07499999999999998,
"neural_mlp": 0.06545454545454546,
"better_baseline": "minimal"
}
},
"next_subtask_forecast": {
"name": "Long-horizon next-subtask forecasting",
"family": "classification",
"input": "current and historical windows",
"output": "future procedure-step label",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"D": "proxy"
},
"why": "Measures whether the model can anticipate the next procedural phase rather than only the current frame state.",
"current_limit": "Subtask labels are constrained to the available annotation vocabulary.",
"display_name": "Long-Horizon Next-Subtask Forecasting",
"artifact_id": "next_subtask_forecast",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 0.04545454545454545,
"neural_mlp": 0.050724637681159424,
"better_baseline": "neural_mlp"
}
},
"interaction_text_prediction": {
"name": "Interaction text prediction",
"family": "classification",
"input": "window features without target text leakage",
"output": "natural-language interaction class",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"A": "proxy"
},
"why": "Connects egocentric observations to the natural-language interaction semantics carried by the annotation.",
"current_limit": "Public derived features retain hashed text targets; raw full text requires the official annotation source.",
"display_name": "Interaction Text Prediction",
"artifact_id": "interaction_text_prediction",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 0.04444444444444444,
"neural_mlp": 0.0380952380952381,
"better_baseline": "minimal"
}
},
"action_object_relation": {
"name": "Action-object relation prediction",
"family": "classification",
"input": "window features with target-side relation leakage excluded",
"output": "action-object relation class",
"primary_direction": "C",
"direction_roles": {
"C": "direct",
"D": "proxy"
},
"why": "Tests whether action recognition and object state are connected as a relational interaction representation.",
"current_limit": "Relation labels are derived from the public-sample annotation scope.",
"display_name": "Action-Object Relation Prediction",
"artifact_id": "action_object_relation",
"metric": {
"key": "macro_f1",
"name": "macro-F1",
"direction": "higher",
"minimal": 0.0,
"neural_mlp": 0.0,
"better_baseline": "tie"
}
},
"object_set_forecast": {
"name": "Future object-set forecasting",
"family": "multi-label",
"input": "current and historical windows",
"output": "future object set",
"primary_direction": "D",
"direction_roles": {
"D": "direct",
"C": "proxy"
},
"why": "Asks whether the current scene state supports predicting which objects will matter later.",
"current_limit": "This is a set-level proxy, not a persistent 3D scene graph.",
"display_name": "Future Object-Set Forecasting",
"artifact_id": "object_set_forecast",
"metric": {
"key": "micro_f1",
"name": "micro-F1",
"direction": "higher",
"minimal": 0.16939890710382516,
"neural_mlp": 0.19718309859154928,
"better_baseline": "neural_mlp"
}
},
"imu_to_hand_pose": {
"name": "IMU-to-hand pose reconstruction",
"family": "regression",
"input": "IMU and motion context",
"output": "hand pose target",
"primary_direction": "A",
"direction_roles": {
"A": "direct",
"B": "proxy"
},
"why": "Measures human-motion reconstruction from wearable and motion cues.",
"current_limit": "Pose reconstruction is window-level and does not yet fit a full parametric hand/body model.",
"display_name": "IMU-to-Hand Pose Reconstruction",
"artifact_id": "imu_to_hand_pose",
"metric": {
"key": "mae",
"name": "MAE",
"direction": "lower",
"minimal": 0.042049407958984375,
"neural_mlp": 0.042562149465084076,
"better_baseline": "minimal"
}
},
"camera_view_sync_retrieval": {
"name": "Camera-view synchronization retrieval",
"family": "retrieval",
"input": "one camera-view/window query",
"output": "matching synchronized view",
"primary_direction": "B",
"direction_roles": {
"B": "direct",
"D": "proxy"
},
"why": "Tests whether synchronized multi-view structure is recoverable across camera streams.",
"current_limit": "Retrieval checks view consistency but does not reconstruct geometry by itself.",
"display_name": "Camera-View Synchronization Retrieval",
"artifact_id": "camera_view_sync_retrieval",
"metric": {
"key": "mrr",
"name": "MRR",
"direction": "higher",
"minimal": 0.4943004846572876,
"neural_mlp": 0.24086658656597137,
"better_baseline": "minimal"
}
},
"time_to_transition": {
"name": "Time-to-next-transition regression",
"family": "regression",
"input": "current temporal window state",
"output": "frames/time until the next transition",
"primary_direction": "C",
"direction_roles": {
"C": "diagnostic",
"D": "diagnostic"
},
"why": "Measures temporal boundary awareness as a continuous timing target.",
"current_limit": "Regression is local to the annotated public sample timeline.",
"display_name": "Time-to-Next-Transition Regression",
"artifact_id": "time_to_transition",
"metric": {
"key": "mae",
"name": "MAE frames",
"direction": "lower",
"minimal": 10.53735637664795,
"neural_mlp": 10.55449390411377,
"better_baseline": "minimal"
}
}
}
}
|