Spaces:
Sleeping
Sleeping
File size: 4,544 Bytes
44e5f85 9765f33 44e5f85 | 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 |
// Axis vocabulary transcribed from T2AV_guideline_final_updated.html (section 02,
// "The six evaluation axes" and section 04 "Axis-by-axis rules"). This is the
// single source other components pull axis copy from.
const axisDefinitions = {
a: {
id: "a",
key: "A",
title: "Audio Event Correctness",
granularity: "per-event",
summary: "Whether the required speech or sound event is present and correct in the audio.",
detail:
"PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families."
},
v: {
id: "v",
key: "V",
title: "Visual Event Correctness",
granularity: "per-event",
summary:
"Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.",
detail:
"PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance."
},
avb: {
id: "avb",
key: "AVB",
title: "Audio-Visual Binding",
granularity: "per-event",
summary: "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.",
detail:
"Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB."
},
tcRel: {
id: "tcRel",
key: "TC_REL",
title: "Temporal Relation",
granularity: "per-video",
summary: "Whether the specified relationship between events holds - synchronization, order, and absolute timing.",
detail:
"Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line."
},
tcCount: {
id: "tcCount",
key: "TC_COUNT",
title: "Temporal Count",
granularity: "per-event",
summary: "Whether each required event occurs the specified number of times, with no missing or extra occurrences.",
detail:
"Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another."
},
cc: {
id: "cc",
key: "CC",
title: "Constraint Compliance",
granularity: "per-video",
summary:
"Non-core constraints: scene, camera, subject count, no subtitles/BGM, no extra actions or sounds, post-event stillness when required.",
detail:
"Everything not already covered above: scene and camera conditions; number and identity of subjects; initial pose and required visibility; absence of subtitles, on-screen text, or background music; absence of additional deliberate actions, speech, or salient sounds; and post-event stillness only when explicitly required. Allowed natural motion: blinking, breathing, minor posture adjustment, necessary preparation, and natural reset motion."
}
};
export const hierarchyFlow = "A / V -> AVB -> TC_REL and TC_COUNT -> CC";
export const recordingOrder = [
"Identify each required event from the prompt.",
"Evaluate A and V (per event).",
"Apply the hierarchy -> AVB (per audio+visual event).",
"Evaluate TC_REL, then TC_COUNT.",
"Evaluate CC for remaining constraints and any unrelated extra events."
];
export default axisDefinitions;
|