File size: 15,413 Bytes
a32bc15
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
 "name": "VisionQ six-axis vocabulary (6 axes, 51 leaves)",
 "kind": "root",
 "children": [
  {
   "name": "Image Appearance",
   "kind": "top_cluster",
   "definition": "Whole-panel, crop-level, or panel-level low-level visual appearance.",
   "evaluator_question": "Does the figure, crop, or panel have acceptable low-level visual appearance?",
   "positive_scope": "Blur, noise, exposure, contrast, color, lighting, global artifact, realism, and style.",
   "negative_scope": "Object shape errors, spatial relation errors, prompt mismatch, reference mismatch, point distribution, keypoint accuracy, mask accuracy, and latent attention-map behavior.",
   "children": [
    {
     "name": "Blur",
     "kind": "l1_cluster",
     "definition": "Focus loss, smear, or blur visible at the panel or crop level.",
     "children": []
    },
    {
     "name": "Noise",
     "kind": "l1_cluster",
     "definition": "Random speckle, grain, sensor-like corruption, or noisy visual artifacts.",
     "children": []
    },
    {
     "name": "Exposure",
     "kind": "l1_cluster",
     "definition": "Over-exposure, under-exposure, saturation clipping, or visibility loss from brightness.",
     "children": []
    },
    {
     "name": "Contrast",
     "kind": "l1_cluster",
     "definition": "Insufficient or excessive tonal separation that harms visible interpretation.",
     "children": []
    },
    {
     "name": "Color",
     "kind": "l1_cluster",
     "definition": "Color cast, color inconsistency, or visually implausible color independent of prompt/reference criteria.",
     "children": []
    },
    {
     "name": "Lighting",
     "kind": "l1_cluster",
     "definition": "Illumination, shadow, reflection, or lighting quality visible in the output.",
     "children": []
    },
    {
     "name": "Global Artifact",
     "kind": "l1_cluster",
     "definition": "Panel-wide artifact not better explained by object form, layout, relation, prompt, or reference match.",
     "children": []
    },
    {
     "name": "Realism",
     "kind": "l1_cluster",
     "definition": "Overall perceptual plausibility or photorealism when not tied to a specific object, prompt, or reference.",
     "children": []
    },
    {
     "name": "Style",
     "kind": "l1_cluster",
     "definition": "Global visual style or aesthetic quality when judged directly rather than as reference or prompt match.",
     "children": []
    }
   ]
  },
  {
   "name": "Object Form",
   "kind": "top_cluster",
   "definition": "Internal quality of visible objects, regions, bodies, faces, surfaces, and masks.",
   "evaluator_question": "Are visible objects or regions internally well-formed?",
   "positive_scope": "Shape, boundary, texture, surface, material, completeness, pose, face, anatomy, detail, and segmentation boundary quality.",
   "negative_scope": "Whole-scene arrangement, pairwise relation, explicit prompt matching, and visible reference or ground-truth matching.",
   "children": [
    {
     "name": "Shape",
     "kind": "l1_cluster",
     "definition": "Object geometry, silhouette, deformation, or structural shape correctness.",
     "children": []
    },
    {
     "name": "Boundary",
     "kind": "l1_cluster",
     "definition": "Object or region edge quality, contour sharpness, silhouette completeness, or boundary artifacts.",
     "children": []
    },
    {
     "name": "Texture",
     "kind": "l1_cluster",
     "definition": "Local texture quality, texture continuity, or texture corruption on visible regions.",
     "children": []
    },
    {
     "name": "Surface",
     "kind": "l1_cluster",
     "definition": "Surface smoothness, topology, wrinkles, holes, or visible surface defects.",
     "children": []
    },
    {
     "name": "Material",
     "kind": "l1_cluster",
     "definition": "Material appearance or material property quality of an object or region.",
     "children": []
    },
    {
     "name": "Completeness",
     "kind": "l1_cluster",
     "definition": "Missing, truncated, duplicated, or incomplete object parts.",
     "children": []
    },
    {
     "name": "Pose",
     "kind": "l1_cluster",
     "definition": "Pose plausibility of a single visible object, body, or articulated entity when not judged against ground truth.",
     "children": []
    },
    {
     "name": "Face",
     "kind": "l1_cluster",
     "definition": "Face-specific form quality including expression quality when not prompt- or reference-driven.",
     "children": []
    },
    {
     "name": "Anatomy",
     "kind": "l1_cluster",
     "definition": "Body, limb, hand, joint, or anatomical plausibility of one entity.",
     "children": []
    },
    {
     "name": "Detail",
     "kind": "l1_cluster",
     "definition": "Fine detail, thin structures, small parts, or local feature preservation.",
     "children": []
    },
    {
     "name": "Segmentation",
     "kind": "l1_cluster",
     "definition": "Mask or segmentation visual quality when judged as object-region form rather than GT match.",
     "children": []
    }
   ]
  },
  {
   "name": "Scene Layout",
   "kind": "top_cluster",
   "definition": "Scene-level spatial organization and physical arrangement.",
   "evaluator_question": "Are scene-level spatial properties correct?",
   "positive_scope": "Composition, depth, scale, view, spatial distribution, occlusion layout, camera geometry, and physical plausibility.",
   "negative_scope": "Object-internal defects, exact pairwise relation claims, explicit reference matching, and prompt-specific layout instructions.",
   "children": [
    {
     "name": "Composition",
     "kind": "l1_cluster",
     "definition": "Overall arrangement of regions, objects, foreground, background, or panel contents.",
     "children": []
    },
    {
     "name": "Depth",
     "kind": "l1_cluster",
     "definition": "Depth ordering, 3D arrangement, or perceived scene depth.",
     "children": []
    },
    {
     "name": "Scale",
     "kind": "l1_cluster",
     "definition": "Relative or absolute size plausibility at scene level.",
     "children": []
    },
    {
     "name": "View",
     "kind": "l1_cluster",
     "definition": "Viewpoint, novel-view layout, or multi-view scene-level consistency.",
     "children": []
    },
    {
     "name": "Spatial Distribution",
     "kind": "l1_cluster",
     "definition": "Distribution, spacing, clumping, or spread of visible scene elements or point samples.",
     "children": []
    },
    {
     "name": "Occlusion Layout",
     "kind": "l1_cluster",
     "definition": "Occlusion ordering or impossible scene-level overlap patterns.",
     "children": []
    },
    {
     "name": "Camera Geometry",
     "kind": "l1_cluster",
     "definition": "Camera perspective, projection geometry, or viewpoint geometry visible in the figure.",
     "children": []
    },
    {
     "name": "Physical Plausibility",
     "kind": "l1_cluster",
     "definition": "Scene-level physical plausibility such as floating objects, floor contact at scene scale, or impossible arrangements.",
     "children": []
    }
   ]
  },
  {
   "name": "Relation",
   "kind": "top_cluster",
   "definition": "Correctness of relationships between visible entities, parts, attributes, or correspondence points.",
   "evaluator_question": "Are relationships between visible entities correct?",
   "positive_scope": "Contact, correspondence, attribute binding, spatial relation, interaction, and part-whole relation.",
   "negative_scope": "Single-entity pose plausibility, GT/keypoint accuracy, bounding-box accuracy, segmentation quality, and protocol robustness.",
   "children": [
    {
     "name": "Contact",
     "kind": "l1_cluster",
     "definition": "Whether two visible entities or parts touch, connect, intersect, or maintain contact correctly.",
     "children": []
    },
    {
     "name": "Correspondence",
     "kind": "l1_cluster",
     "definition": "Visible correspondence between entities, parts, points, views, or panels.",
     "children": []
    },
    {
     "name": "Attribute Binding",
     "kind": "l1_cluster",
     "definition": "Correct binding of visible attributes to the right entity or part.",
     "children": []
    },
    {
     "name": "Spatial Relation",
     "kind": "l1_cluster",
     "definition": "Pairwise relation such as left/right, above/below, inside/outside, near/far, or in-front/behind.",
     "children": []
    },
    {
     "name": "Interaction",
     "kind": "l1_cluster",
     "definition": "Visible interaction between people, objects, agents, tools, or scene elements.",
     "children": []
    },
    {
     "name": "Part-Whole Relation",
     "kind": "l1_cluster",
     "definition": "Relationship between an object and its visible parts when the relation itself is the criterion.",
     "children": []
    }
   ]
  },
  {
   "name": "Reference Match",
   "kind": "top_cluster",
   "definition": "Fidelity to a visible source, target, ground truth, before-after panel, or comparison reference.",
   "evaluator_question": "Does the output match a visible source, target, ground truth, or comparison panel?",
   "positive_scope": "Identity, background, reconstruction, edit preservation, geometry match, style match, mask match, landmark match, and segmentation match.",
   "negative_scope": "Prompt-only matching, absolute image appearance with no reference, single-object form with no visible reference, and protocol robustness.",
   "children": [
    {
     "name": "Identity",
     "kind": "l1_cluster",
     "definition": "Identity preservation or identity match against a visible source/reference.",
     "children": []
    },
    {
     "name": "Background",
     "kind": "l1_cluster",
     "definition": "Background preservation or match to visible source/reference.",
     "children": []
    },
    {
     "name": "Reconstruction",
     "kind": "l1_cluster",
     "definition": "Reconstruction fidelity against target, source, or ground truth.",
     "children": []
    },
    {
     "name": "Edit Preservation",
     "kind": "l1_cluster",
     "definition": "Preservation of non-edited content in before/after or edit-reference setups.",
     "children": []
    },
    {
     "name": "Geometry Match",
     "kind": "l1_cluster",
     "definition": "Geometric fidelity to visible source, target, annotation, or ground truth.",
     "children": []
    },
    {
     "name": "Style Match",
     "kind": "l1_cluster",
     "definition": "Style fidelity when a visible reference style is the criterion.",
     "children": []
    },
    {
     "name": "Mask Match",
     "kind": "l1_cluster",
     "definition": "Mask fidelity against visible mask, annotation, or ground truth.",
     "children": []
    },
    {
     "name": "Landmark Match",
     "kind": "l1_cluster",
     "definition": "Keypoint, landmark, or pose-coordinate match against visible ground truth or annotation.",
     "children": []
    },
    {
     "name": "Segmentation Match",
     "kind": "l1_cluster",
     "definition": "Segmentation output fidelity against visible ground-truth segmentation.",
     "children": []
    }
   ]
  },
  {
   "name": "Prompt Match",
   "kind": "top_cluster",
   "definition": "Match to a text prompt, caption, instruction, requested content, or requested attribute.",
   "evaluator_question": "Does the output match the text prompt, caption, or instruction?",
   "positive_scope": "Object presence, attribute match, spatial instruction, semantic match, text legibility, count match, action match, and expression match.",
   "negative_scope": "Lighting, expression, physical plausibility, or object quality when not specified by prompt/caption/instruction.",
   "children": [
    {
     "name": "Object Presence",
     "kind": "l1_cluster",
     "definition": "Requested object, entity, category, or scene element is present or absent as specified.",
     "children": []
    },
    {
     "name": "Attribute Match",
     "kind": "l1_cluster",
     "definition": "Requested color, material, style, state, or other attribute is correctly assigned.",
     "children": []
    },
    {
     "name": "Spatial Instruction",
     "kind": "l1_cluster",
     "definition": "Prompt-specified spatial arrangement or relation is satisfied.",
     "children": []
    },
    {
     "name": "Semantic Match",
     "kind": "l1_cluster",
     "definition": "Overall semantic content matches the caption, prompt, or instruction.",
     "children": []
    },
    {
     "name": "Text Legibility",
     "kind": "l1_cluster",
     "definition": "Requested text, glyphs, symbols, or readable writing are rendered correctly.",
     "children": []
    },
    {
     "name": "Count Match",
     "kind": "l1_cluster",
     "definition": "Requested number of objects or entities is correct.",
     "children": []
    },
    {
     "name": "Action Match",
     "kind": "l1_cluster",
     "definition": "Requested action, activity, or event is depicted correctly.",
     "children": []
    },
    {
     "name": "Expression Match",
     "kind": "l1_cluster",
     "definition": "Requested facial expression, emotion, or pose state is depicted correctly.",
     "children": []
    }
   ]
  }
 ],
 "sidecar": {
  "name": "Sidecar / Rejected",
  "kind": "sidecar",
  "definition": "Research concepts that should not be forced into the main qualitative figure tree.",
  "reason_codes": [
   "temporal_out_of_scope",
   "method_or_nonvisual_property",
   "protocol_robustness",
   "latent_internal_metric",
   "dataset_training_property",
   "metric_calibration",
   "multi_sample_required",
   "not_visible_in_figure",
   "requires_external_knowledge"
  ],
  "children": [
   {
    "name": "Temporal Behavior",
    "kind": "sidecar_cluster",
    "definition": "Claims requiring video playback, sequence behavior, flicker, jitter, or temporal dynamics."
   },
   {
    "name": "Multi-Sample Diversity",
    "kind": "sidecar_cluster",
    "definition": "Claims requiring multiple generated samples or distribution-level diversity evidence."
   },
   {
    "name": "Method Control",
    "kind": "sidecar_cluster",
    "definition": "Claims about controllability or intervention behavior not directly judged from a single qualitative figure."
   },
   {
    "name": "Protocol Robustness",
    "kind": "sidecar_cluster",
    "definition": "Claims about robustness to perturbations, occlusion protocols, or test-time conditions."
   },
   {
    "name": "Latent Representation",
    "kind": "sidecar_cluster",
    "definition": "Claims about attention maps, internal features, embeddings, or hidden model states when not used as visible figure output."
   },
   {
    "name": "Dataset Coverage",
    "kind": "sidecar_cluster",
    "definition": "Claims about dataset distribution, coverage, domain breadth, or benchmark composition."
   },
   {
    "name": "Training Memorization",
    "kind": "sidecar_cluster",
    "definition": "Claims about memorization, training data leakage, or learned representations."
   },
   {
    "name": "Metric Calibration",
    "kind": "sidecar_cluster",
    "definition": "Claims about numerical metrics, calibration, FID, LPIPS, IoU, or other score behavior."
   },
   {
    "name": "Nonvisual Method Property",
    "kind": "sidecar_cluster",
    "definition": "Claims about speed, efficiency, architecture, optimization, or generalization without direct visual figure evidence."
   }
  ]
 }
}