Train30 · Caption + Actions + Temporal Localization

Purple highlights mark action phrases; entity underline colors correspond to the entity-marked frames. Toggle between clean and entity-marked frames.

30 videos / 30 model calls (cached)

77 actions

129/129 linked subject/object entities

30/30 text-position checks passed

77/77 localized action intervals · 30 model calls

Text checks validate character offsets and entity-mark references; expand each action to inspect its temporal intervals.

View unified prompt · View action prompt v2 · Download full results · View run statistics

View temporal-localization prompt · Each action shows its intervals and boundary frames. "View segment" highlights the corresponding clean frames above.

0_wHveSGjXyDY

4 actionsText check: ok9.2s · 1 model call

Frame 10 (f9) · source frame 232 · 4.40s

Frame 11 (f10) · source frame 241 · 5.00s

Frame 12 (f11) · source frame 247 · 5.40s

Frame 13 (f12) · source frame 256 · 6.00s

Frame 14 (f13) · source frame 262 · 6.40s

Frame 15 (f14) · source frame 271 · 7.00s

Frame 16 (f15) · source frame 277 · 7.40s

Frame 17 (f16) · source frame 286 · 8.00s

Frame 18 (f17) · source frame 292 · 8.40s

Frame 19 (f18) · source frame 301 · 9.00s

Caption + Actions + Temporal Localization

An older man repeatedly strikes a metal assembly. He straightens as a red-clad woman watches. Two other men stand nearby.

strikessimple · highSubject: An older man g0 → #4Object: a metal assembly g1 → #2

Located · Localization confidence medium: The older man repeatedly raises and lowers his striking arm toward the assembly.

Frame sequence:[[7, 13]]Time sequence:[[3.4, 6.9]]

Frames 8–14 · 3.40–6.90s (end exclusive)

Code indices f7–f13 (0-based, 2 FPS sampled sequence); source files 00000217.jpg → 00000262.jpg. Last observed frame time: 6.40s.

strikes first action frame — First observed frame · Frame 8

strikes last action frame — Last observed frame · Frame 14

Interval boundary estimated by the model from saved frames, not human ground truth.

straightenssimple · highSubject: He g0 → #4

Located · Localization confidence high: He raises his torso from the deeply bent working position.

Frame sequence:[[14, 17]]Time sequence:[[7.0, 8.9]]

Frames 15–18 · 7.00–8.90s (end exclusive)

Code indices f14–f17 (0-based, 2 FPS sampled sequence); source files 00000271.jpg → 00000292.jpg. Last observed frame time: 8.40s.

straightens first action frame — First observed frame · Frame 15

straightens last action frame — Last observed frame · Frame 18

Interval boundary estimated by the model from saved frames, not human ground truth.

watchessimple · highSubject: a red-clad woman g2 → #5

Located · Localization confidence medium: The woman remains behind the assembly, looking toward the man's work.

Frame sequence:[[0, 18]]Time sequence:[[0.0, 9.2]]

Frames 1–19 · 0.00–9.20s (end exclusive)

Code indices f0–f18 (0-based, 2 FPS sampled sequence); source files 00000166.jpg → 00000301.jpg. Last observed frame time: 9.00s.

watches first action frame — First observed frame · Frame 1

watches last action frame — Last observed frame · Frame 19

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

standposture · highSubject: Two other men g3 → #3, g4 → #6

Located · Localization confidence medium: Both men hold standing positions beside the assembly during these intervals; the left man steps between them.

Frame sequence:[[9, 13], [16, 18]]Time sequence:[[4.4, 6.9], [8.0, 9.2]]

Frames 10–14 · 4.40–6.90s (end exclusive)

Code indices f9–f13 (0-based, 2 FPS sampled sequence); source files 00000232.jpg → 00000262.jpg. Last observed frame time: 6.40s.

stand first action frame — First observed frame · Frame 10

stand last action frame — Last observed frame · Frame 14

Interval boundary estimated by the model from saved frames, not human ground truth.

Frames 17–19 · 8.00–9.20s (end exclusive)

Code indices f16–f18 (0-based, 2 FPS sampled sequence); source files 00000286.jpg → 00000301.jpg. Last observed frame time: 9.00s.

Clip end reached; the true end may be later

g0 → #4 · person
An older man / He Category check: agreeg1 → #2 · other machine
a metal assembly Category check: unknowng2 → #5 · person
a red-clad woman Category check: agreeg3 → #3 · person
Two other men Category check: agreeg4 → #6 · person
Two other men Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 5/7; subject/object links: 5; unlinked: 0.

1093_TZUYCiN-7T0

2 actionsText check: ok3.0s · 1 model call

Caption + Actions + Temporal Localization

Two dogs run side by side across the snow. They kick up snow behind them.

runsimple · highSubject: Two dogs g0 → #4, g1 → #5

Located · Localization confidence high: Both dogs show changing running strides as they advance together throughout the sequence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00000828.jpg → 00000870.jpg. Last observed frame time: 2.80s.

run first action frame — First observed frame · Frame 1

run last action frame — Last observed frame · Frame 7

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

kick upphrasal · highSubject: They g0 → #4, g1 → #5Object: snow g2 → #2

Located · Localization confidence medium: Disturbed snow sprays around and behind the dogs' feet as they run.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00000828.jpg → 00000870.jpg. Last observed frame time: 2.80s.

kick up first action frame — First observed frame · Frame 1

kick up last action frame — Last observed frame · Frame 7

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #4 · dog
Two dogs / They / them Category check: agreeg1 → #5 · dog
Two dogs / They / them Category check: agreeg2 → #2 · snowfield
the snow / snow Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 3/5; subject/object links: 3; unlinked: 0.

1182_tNbLX2wX4Bs

3 actionsText check: ok9.2s · 1 model call

Frame 10 (f9) · source frame 818 · 4.40s

Frame 11 (f10) · source frame 827 · 5.00s

Frame 12 (f11) · source frame 833 · 5.40s

Frame 13 (f12) · source frame 842 · 6.00s

Frame 14 (f13) · source frame 848 · 6.40s

Frame 15 (f14) · source frame 857 · 7.00s

Frame 16 (f15) · source frame 863 · 7.40s

Frame 17 (f16) · source frame 872 · 8.00s

Frame 18 (f17) · source frame 878 · 8.40s

Frame 19 (f18) · source frame 887 · 9.00s

Caption + Actions + Temporal Localization

Many children stand in a circle indoors beside two women. A boy steps in place among them. One woman gestures toward the children.

standposture · highSubject: Many children g0 → #21, g1 → #17, g2 → #23, g3 → #18, g4 → #14, g5 → #11, g6 → #9, g7 → #8, g8 → #24, g9 → #7, g10 → #27, g11 → #13, g12 → #16, g13 → #20, g14 → #22, g15 → #25, g16 → #15, g17 → #38, g18 → #12, g19 → #26, g20 → #39Object: two women g21 → #10, g22 → #19

Located · Localization confidence medium: The children remain upright in a circle throughout, although several are partly obscured.

Frame sequence:[[0, 18]]Time sequence:[[0.0, 9.2]]

Frames 1–19 · 0.00–9.20s (end exclusive)

Code indices f0–f18 (0-based, 2 FPS sampled sequence); source files 00000752.jpg → 00000887.jpg. Last observed frame time: 9.00s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

stepssimple · highSubject: A boy g1 → #17Object: them g0 → #21, g1 → #17, g2 → #23, g3 → #18, g4 → #14, g5 → #11, g6 → #9, g7 → #8, g8 → #24, g9 → #7, g10 → #27, g11 → #13, g12 → #16, g13 → #20, g14 → #22, g15 → #25, g16 → #15, g17 → #38, g18 → #12, g19 → #26, g20 → #39

Located · Localization confidence high: The central boy repeatedly lifts and lowers his feet in place before settling into a stationary stance.

Frame sequence:[[0, 9]]Time sequence:[[0.0, 4.9]]

Frames 1–10 · 0.00–4.90s (end exclusive)

Code indices f0–f9 (0-based, 2 FPS sampled sequence); source files 00000752.jpg → 00000818.jpg. Last observed frame time: 4.40s.

steps first action frame — First observed frame · Frame 1

steps last action frame — Last observed frame · Frame 10

Clip start reached; the true start may be earlier

gesturessimple · highSubject: One woman g21 → #10Object: the children g0 → #21, g1 → #17, g2 → #23, g3 → #18, g4 → #14, g5 → #11, g6 → #9, g7 → #8, g8 → #24, g9 → #7, g10 → #27, g11 → #13, g12 → #16, g13 → #20, g14 → #22, g15 → #25, g16 → #15, g17 → #38, g18 → #12, g19 → #26, g20 → #39

Located · Localization confidence high: The woman on the left extends a hand toward the circle initially and raises it again near the end.

Frame sequence:[[0, 1], [15, 17]]Time sequence:[[0.0, 0.9], [7.4, 8.9]]

Frames 1–2 · 0.00–0.90s (end exclusive)

Code indices f0–f1 (0-based, 2 FPS sampled sequence); source files 00000752.jpg → 00000758.jpg. Last observed frame time: 0.40s.

gestures first action frame — First observed frame · Frame 1

gestures last action frame — Last observed frame · Frame 2

Clip start reached; the true start may be earlier

Frames 16–18 · 7.40–8.90s (end exclusive)

Code indices f15–f17 (0-based, 2 FPS sampled sequence); source files 00000863.jpg → 00000878.jpg. Last observed frame time: 8.40s.

Interval boundary estimated by the model from saved frames, not human ground truth.

g0 → #21 · person
Many children / them / the children Category check: agreeg1 → #17 · person
Many children / A boy / them / the children Category check: agreeg2 → #23 · person
Many children / them / the children Category check: agreeg3 → #18 · person
Many children / them / the children Category check: agreeg4 → #14 · person
Many children / them / the children Category check: agreeg5 → #11 · person
Many children / them / the children Category check: agreeg6 → #9 · person
Many children / them / the children Category check: agreeg7 → #8 · person
Many children / them / the children Category check: agreeg8 → #24 · person
Many children / them / the children Category check: agreeg9 → #7 · person
Many children / them / the children Category check: agreeg10 → #27 · person
Many children / them / the children Category check: agreeg11 → #13 · person
Many children / them / the children Category check: agreeg12 → #16 · person
Many children / them / the children Category check: agreeg13 → #20 · person
Many children / them / the children Category check: agreeg14 → #22 · person
Many children / them / the children Category check: agreeg15 → #25 · person
Many children / them / the children Category check: agreeg16 → #15 · person
Many children / them / the children Category check: agreeg17 → #38 · person
Many children / them / the children Category check: agreeg18 → #12 · person
Many children / them / the children Category check: agreeg19 → #26 · person
Many children / them / the children Category check: agreeg20 → #39 · person
Many children / them / the children Category check: agreeg21 → #10 · person
two women / One woman Category check: agreeg22 → #19 · person
two women Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 23/39; subject/object links: 6; unlinked: 0.

1270_9oDHw4wwTQI

4 actionsText check: ok3.0s · 1 model call

Caption + Actions + Temporal Localization

A boy and a girl dance on the turquoise floor. They turn together. Many spectators watch nearby, and some hold video cameras.

dancesimple · highSubject: A boy g0 → #14Object: a girl g1 → #15

Located · Localization confidence high: The boy and girl move together in a partnered dance hold throughout the sequence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00000715.jpg → 00000757.jpg. Last observed frame time: 2.80s.

dance first action frame — First observed frame · Frame 1

dance last action frame — Last observed frame · Frame 7

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

turnsimple · highSubject: They g0 → #14, g1 → #15

Located · Localization confidence high: The pair progressively changes orientation while maintaining their dance hold.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00000715.jpg → 00000757.jpg. Last observed frame time: 2.80s.

turn first action frame — First observed frame · Frame 1

turn last action frame — Last observed frame · Frame 7

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

watchsimple · mediumSubject: Many spectators g3 → #13, g4 → #12, g5 → #25, g6 → #26, g7 → #29, g8 → #19, g9 → #30, g10 → #18, g11 → #32, g12 → #33, g13 → #20, g14 → #37, g15 → #31, g16 → #38, g17 → #21, g18 → #22

Located · Localization confidence medium: Spectators face the dancing area throughout, although different members are visible at different times.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00000715.jpg → 00000757.jpg. Last observed frame time: 2.80s.

watch first action frame — First observed frame · Frame 1

watch last action frame — Last observed frame · Frame 7

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

holdsimple · highSubject: some g3 → #13, g4 → #12, g7 → #29, g8 → #19Object: video cameras g19 → #3

Located · Localization confidence high: Members of the referenced subset visibly hold cameras toward the dancers across the sequence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00000715.jpg → 00000757.jpg. Last observed frame time: 2.80s.

hold first action frame — First observed frame · Frame 1

hold last action frame — Last observed frame · Frame 7

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #14 · person
A boy / They Category check: agreeg1 → #15 · person
a girl / They Category check: agreeg2 → #2 · floor
the turquoise floor Category check: agreeg3 → #13 · person
Many spectators / some Category check: agreeg4 → #12 · person
Many spectators / some Category check: agreeg5 → #25 · person
Many spectators Category check: agreeg6 → #26 · person
Many spectators Category check: agreeg7 → #29 · person
Many spectators / some Category check: agreeg8 → #19 · person
Many spectators / some Category check: agreeg9 → #30 · person
Many spectators Category check: agreeg10 → #18 · person
Many spectators Category check: agreeg11 → #32 · person
Many spectators Category check: agreeg12 → #33 · person
Many spectators Category check: agreeg13 → #20 · person
Many spectators Category check: agreeg14 → #37 · person
Many spectators Category check: agreeg15 → #31 · person
Many spectators Category check: agreeg16 → #38 · person
Many spectators Category check: agreeg17 → #21 · person
Many spectators Category check: agreeg18 → #22 · person
Many spectators Category check: agreeg19 → #3 · other electronic product
video cameras Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 20/40; subject/object links: 6; unlinked: 0.

1363_QwwaDqYgOYo

2 actionsText check: ok3.0s · 1 model call

Frame 1 (f0) · source frame 2448 · 0.00s

Frame 2 (f1) · source frame 2454 · 0.40s

Frame 3 (f2) · source frame 2463 · 1.00s

Frame 4 (f3) · source frame 2469 · 1.40s

Frame 5 (f4) · source frame 2478 · 2.00s

Frame 6 (f5) · source frame 2484 · 2.40s

Frame 7 (f6) · source frame 2490 · 2.80s

Caption + Actions + Temporal Localization

A black-and-white dog wriggles against a leopard-print blanket on a bed. It then lies stretched out beside the blanket.

wrigglessimple · mediumSubject: A black-and-white dog g0 → #5Object: a leopard-print blanket g1 → #3Object: a bed g2 → #6

Located · Localization confidence medium: The dog shifts its head and body against the blanket during the early frames.

Frame sequence:[[0, 3]]Time sequence:[[0.0, 1.9]]

Frames 1–4 · 0.00–1.90s (end exclusive)

Code indices f0–f3 (0-based, 2 FPS sampled sequence); source files 00002448.jpg → 00002469.jpg. Last observed frame time: 1.40s.

wriggles first action frame — First observed frame · Frame 1

wriggles last action frame — Last observed frame · Frame 4

Clip start reached; the true start may be earlier

liesposture · highSubject: It g0 → #5Object: the blanket g1 → #3

Located · Localization confidence high: The dog remains stretched out with its belly down and hind legs extended in the later frames.

Frame sequence:[[3, 6]]Time sequence:[[1.4, 3.0]]

Frames 4–7 · 1.40–3.00s (end exclusive)

Code indices f3–f6 (0-based, 2 FPS sampled sequence); source files 00002469.jpg → 00002490.jpg. Last observed frame time: 2.80s.

lies first action frame — First observed frame · Frame 4

lies last action frame — Last observed frame · Frame 7

Clip end reached; the true end may be later

g0 → #5 · dog
A black-and-white dog / It Category check: agreeg1 → #3 · textiles
a leopard-print blanket / the blanket Category check: agreeg2 → #6 · bed
a bed Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 3/8; subject/object links: 5; unlinked: 0.

1529_N6hl1No1Syc

0 actionsText check: ok3.0s · 1 model call

Caption + Actions + Temporal Localization

A dark wooden chair and a lighter chair are on bare ground among trees. They are beside scattered logs and a pale stone.

No actions extracted.

g0 → #7 · chair or seat
A dark wooden chair / They Category check: agreeg1 → #8 · chair or seat
a lighter chair / They Category check: agreeg2 → #1 · ground
bare ground Category check: agreeg3 → #6 · tree
trees Category check: agreeg4 → #5 · wood
scattered logs Category check: agreeg5 → #4 · stone
a pale stone Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 6/8; subject/object links: 0; unlinked: 0.

1665_7wOrT_feKvA

1 actionsText check: ok3.0s · 1 model call

Frame 1 (f0) · source frame 2628 · 0.00s

Frame 2 (f1) · source frame 2634 · 0.40s

Frame 3 (f2) · source frame 2643 · 1.00s

Frame 4 (f3) · source frame 2649 · 1.40s

Frame 5 (f4) · source frame 2658 · 2.00s

Frame 6 (f5) · source frame 2664 · 2.40s

Frame 7 (f6) · source frame 2670 · 2.80s

Caption + Actions + Temporal Localization

A blue-and-white boat travels along the river beneath a raised bridge. It moves steadily to the right.

travels / movessimple · highSubject: A blue-and-white boat g0 → #8Object: a raised bridge g2 → #4

Located · Localization confidence high: The boat advances steadily rightward beneath the stationary bridge throughout the clean-frame sequence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00002628.jpg → 00002670.jpg. Last observed frame time: 2.80s.

travels first action frame — First observed frame · Frame 1

travels last action frame — Last observed frame · Frame 7

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #8 · ship or boat
A blue-and-white boat / It Category check: agreeg1 → #6 · river
the river Category check: agreeg2 → #4 · bridge
a raised bridge Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 3/8; subject/object links: 2; unlinked: 0.

1804_H89b4qUm51w

2 actionsText check: ok3.0s · 1 model call

Frame 1 (f0) · source frame 3498 · 0.00s

Frame 2 (f1) · source frame 3504 · 0.40s

Frame 3 (f2) · source frame 3513 · 1.00s

Frame 4 (f3) · source frame 3519 · 1.40s

Frame 5 (f4) · source frame 3528 · 2.00s

Frame 6 (f5) · source frame 3534 · 2.40s

Frame 7 (f6) · source frame 3540 · 2.80s

Caption + Actions + Temporal Localization

A woman walks along a stone path toward a wooden planter beside a white house. She briefly reaches toward the leafy plants.

walkssimple · highSubject: A woman g0 → #8Object: a wooden planter g2 → #9

Located · Localization confidence high: She advances along the stone path toward the planter throughout the sequence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00003498.jpg → 00003540.jpg. Last observed frame time: 2.80s.

walks first action frame — First observed frame · Frame 1

walks last action frame — Last observed frame · Frame 7

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

reachesprepositional · mediumSubject: She g0 → #8Object: the leafy plants g4 → #7

Located · Localization confidence medium: She extends her right arm toward the foliage, most clearly in frame 3, then lowers it.

Frame sequence:[[2, 3]]Time sequence:[[1.0, 1.9]]

Frames 3–4 · 1.00–1.90s (end exclusive)

Code indices f2–f3 (0-based, 2 FPS sampled sequence); source files 00003513.jpg → 00003519.jpg. Last observed frame time: 1.40s.

reaches first action frame — First observed frame · Frame 3

reaches last action frame — Last observed frame · Frame 4

Interval boundary estimated by the model from saved frames, not human ground truth.

g0 → #8 · person
A woman / She Category check: agreeg1 → #3 · path
a stone path Category check: agreeg2 → #9 · box
a wooden planter Category check: unknowng3 → #4 · house
a white house Category check: agreeg4 → #7 · other plant
the leafy plants Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 5/9; subject/object links: 4; unlinked: 0.

1959_Ejgrd03o2F8

3 actionsText check: ok3.0s · 1 model call

Frame 1 (f0) · source frame 1668 · 0.00s

Frame 2 (f1) · source frame 1674 · 0.40s

Frame 3 (f2) · source frame 1683 · 1.00s

Frame 4 (f3) · source frame 1689 · 1.40s

Frame 5 (f4) · source frame 1698 · 2.00s

Frame 6 (f5) · source frame 1704 · 2.40s

Frame 7 (f6) · source frame 1710 · 2.80s

Caption + Actions + Temporal Localization

A woman leans on a railing beside two white cabinets. She looks over the water and colorful waterslides. A man in red stands beside the blue slide entrance.

leansposture · highSubject: A woman g0 → #10Object: a railing g1 → #1Object: two white cabinets g2 → #12, g3 → #13

Located · Localization confidence high: The woman maintains a forward lean against the railing until she leaves the visible scene.

Frame sequence:[[0, 3]]Time sequence:[[0.0, 1.9]]

Frames 1–4 · 0.00–1.90s (end exclusive)

Code indices f0–f3 (0-based, 2 FPS sampled sequence); source files 00001668.jpg → 00001689.jpg. Last observed frame time: 1.40s.

leans first action frame — First observed frame · Frame 1

leans last action frame — Last observed frame · Frame 4

Clip start reached; the true start may be earlier

lookssimple · mediumSubject: She g0 → #10Object: the water g4 → #7Object: colorful waterslides g5 → #8

Located · Localization confidence medium: Her head faces outward toward the water and slide area, though her precise gaze target is unclear.

Frame sequence:[[0, 3]]Time sequence:[[0.0, 1.9]]

Frames 1–4 · 0.00–1.90s (end exclusive)

Code indices f0–f3 (0-based, 2 FPS sampled sequence); source files 00001668.jpg → 00001689.jpg. Last observed frame time: 1.40s.

looks first action frame — First observed frame · Frame 1

looks last action frame — Last observed frame · Frame 4

Clip start reached; the true start may be earlier

standsposture · highSubject: A man g6 → #11Object: the blue slide entrance g7 → #4

Located · Localization confidence high: The partially cropped man is upright beside the slide entrance in the first clean frame.

Frame sequence:[[0, 0]]Time sequence:[[0.0, 0.4]]

Frames 1–1 · 0.00–0.40s (end exclusive)

Code indices f0–f0 (0-based, 2 FPS sampled sequence); source files 00001668.jpg → 00001668.jpg. Last observed frame time: 0.00s.

stands first action frame — First observed frame · Frame 1

stands last action frame — Last observed frame · Frame 1

Clip start reached; the true start may be earlier

g0 → #10 · person
A woman / She Category check: agreeg1 → #1 · handrail or fence
a railing Category check: agreeg2 → #12 · box
two white cabinets Category check: disagreeg3 → #13 · box
two white cabinets Category check: disagreeg4 → #7 · water
the water Category check: agreeg5 → #8 · pipeline
colorful waterslides Category check: unknowng6 → #11 · person
A man Category check: agreeg7 → #4 · other construction
the blue slide entrance Category check: disagree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 8/13; subject/object links: 8; unlinked: 0.

2106_m0-AalaDlqU

3 actionsText check: ok2.2s · 1 model call

Frame 1 (f0) · source frame 3258 · 0.00s

Frame 2 (f1) · source frame 3264 · 0.40s

Frame 3 (f2) · source frame 3273 · 1.00s

Frame 4 (f3) · source frame 3279 · 1.40s

Frame 5 (f4) · source frame 3288 · 2.00s

Caption + Actions + Temporal Localization

A boy is seated cross-legged on the patterned floor beside a laptop. He operates it and watches it.

is seatedposture · highSubject: A boy g0 → #5Object: a laptop g2 → #8

Located · Localization confidence high: The boy maintains a cross-legged seated posture beside the laptop throughout the sequence.

Frame sequence:[[0, 4]]Time sequence:[[0.0, 2.2]]

Frames 1–5 · 0.00–2.20s (end exclusive)

Code indices f0–f4 (0-based, 2 FPS sampled sequence); source files 00003258.jpg → 00003288.jpg. Last observed frame time: 2.00s.

is seated first action frame — First observed frame · Frame 1

is seated last action frame — Last observed frame · Frame 5

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

operatessimple · highSubject: He g0 → #5Object: it g2 → #8

Located · Localization confidence medium: His hand remains over the laptop keyboard and changes position slightly as he uses it.

Frame sequence:[[0, 4]]Time sequence:[[0.0, 2.2]]

Frames 1–5 · 0.00–2.20s (end exclusive)

Code indices f0–f4 (0-based, 2 FPS sampled sequence); source files 00003258.jpg → 00003288.jpg. Last observed frame time: 2.00s.

operates first action frame — First observed frame · Frame 1

operates last action frame — Last observed frame · Frame 5

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

watchessimple · highSubject: He g0 → #5Object: it g2 → #8

Located · Localization confidence high: His gaze stays directed toward the laptop across the clean frames.

Frame sequence:[[0, 4]]Time sequence:[[0.0, 2.2]]

Frames 1–5 · 0.00–2.20s (end exclusive)

Code indices f0–f4 (0-based, 2 FPS sampled sequence); source files 00003258.jpg → 00003288.jpg. Last observed frame time: 2.00s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #5 · person
A boy / He Category check: agreeg1 → #2 · floor
the patterned floor Category check: agreeg2 → #8 · computer
a laptop / it Category check: disagree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 3/8; subject/object links: 6; unlinked: 0.

2331_WP9xxG7oYik

2 actionsText check: ok3.0s · 1 model call

Caption + Actions + Temporal Localization

A woman stands in front of a wall of framed posters. She then gestures inward.

standsposture · highSubject: A woman g0 → #6

Located · Localization confidence high: The woman remains upright in the same place throughout the sequence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00000227.jpg → 00000269.jpg. Last observed frame time: 2.80s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

gesturessimple · mediumSubject: She g0 → #6

Located · Localization confidence medium: Her extended arms move inward and slightly downward before holding a closer position.

Frame sequence:[[3, 5]]Time sequence:[[1.4, 2.8]]

Frames 4–6 · 1.40–2.80s (end exclusive)

Code indices f3–f5 (0-based, 2 FPS sampled sequence); source files 00000248.jpg → 00000263.jpg. Last observed frame time: 2.40s.

Interval boundary estimated by the model from saved frames, not human ground truth.

g0 → #6 · person
A woman / She Category check: agreeg1 → #1 · wall
a wall Category check: agreeg2 → #7 · painting or poster
framed posters Category check: agreeg3 → #8 · painting or poster
framed posters Category check: agreeg4 → #9 · painting or poster
framed posters Category check: agreeg5 → #10 · painting or poster
framed posters Category check: agreeg6 → #11 · painting or poster
framed posters Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 7/11; subject/object links: 2; unlinked: 0.

298_ma7hngzCd-I

4 actionsText check: ok9.2s · 1 model call

Frame 1 (f0) · source frame 1337 · 0.00s

Frame 2 (f1) · source frame 1343 · 0.40s

Frame 3 (f2) · source frame 1352 · 1.00s

Frame 4 (f3) · source frame 1358 · 1.40s

Frame 5 (f4) · source frame 1367 · 2.00s

Frame 6 (f5) · source frame 1373 · 2.40s

Frame 7 (f6) · source frame 1382 · 3.00s

Frame 8 (f7) · source frame 1388 · 3.40s

Frame 9 (f8) · source frame 1397 · 4.00s

Frame 10 (f9) · source frame 1403 · 4.40s

Frame 11 (f10) · source frame 1412 · 5.00s

Frame 12 (f11) · source frame 1418 · 5.40s

Frame 13 (f12) · source frame 1427 · 6.00s

Frame 14 (f13) · source frame 1433 · 6.40s

Frame 15 (f14) · source frame 1442 · 7.00s

Frame 16 (f15) · source frame 1448 · 7.40s

Frame 17 (f16) · source frame 1457 · 8.00s

Frame 18 (f17) · source frame 1463 · 8.40s

Frame 19 (f18) · source frame 1472 · 9.00s

Caption + Actions + Temporal Localization

A girl watches many long-snouted fish swim past. She gestures toward them. A large fish glides close by.

watchessimple · highSubject: A girl g0 → #3Object: many long-snouted fish g1 → #29, g2 → #18, g3 → #10, g4 → #7, g5 → #12, g6 → #6, g7 → #8, g8 → #11, g9 → #5, g10 → #4, g11 → #16, g12 → #19, g13 → #15, g14 → #22, g15 → #14, g16 → #21, g17 → #17, g18 → #13, g19 → #25, g20 → #24, g21 → #23, g22 → #31, g23 → #9, g24 → #26, g25 → #30, g26 → #27, g27 → #28, g28 → #32, g29 → #20

Located · Localization confidence high: The girl faces the fish and follows their movement with her gaze.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.4]]

Frames 1–7 · 0.00–3.40s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00001337.jpg → 00001382.jpg. Last observed frame time: 3.00s.

Clip start reached; the true start may be earlier

swimsimple · highSubject: many long-snouted fish g1 → #29, g2 → #18, g3 → #10, g4 → #7, g5 → #12, g6 → #6, g7 → #8, g8 → #11, g9 → #5, g10 → #4, g11 → #16, g12 → #19, g13 → #15, g14 → #22, g15 → #14, g16 → #21, g17 → #17, g18 → #13, g19 → #25, g20 → #24, g21 → #23, g22 → #31, g23 → #9, g24 → #26, g25 → #30, g26 → #27, g27 → #28, g28 → #32, g29 → #20

Located · Localization confidence medium: Fish continually swim through the scene, although individual members appear for different portions of the clip.

Frame sequence:[[0, 18]]Time sequence:[[0.0, 9.2]]

Frames 1–19 · 0.00–9.20s (end exclusive)

Code indices f0–f18 (0-based, 2 FPS sampled sequence); source files 00001337.jpg → 00001472.jpg. Last observed frame time: 9.00s.

swim first action frame — First observed frame · Frame 1

swim last action frame — Last observed frame · Frame 19

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

gesturesprepositional · highSubject: She g0 → #3Object: them g1 → #29, g2 → #18, g3 → #10, g4 → #7, g5 → #12, g6 → #6, g7 → #8, g8 → #11, g9 → #5, g10 → #4, g11 → #16, g12 → #19, g13 → #15, g14 → #22, g15 → #14, g16 → #21, g17 → #17, g18 → #13, g19 → #25, g20 → #24, g21 → #23, g22 → #31, g23 → #9, g24 → #26, g25 → #30, g26 → #27, g27 → #28, g28 → #32, g29 → #20

Located · Localization confidence high: She extends a hand toward the fish in three observed gestures, returning it toward her face between the first two.

Frame sequence:[[5, 5], [7, 7], [14, 14]]Time sequence:[[2.4, 2.9], [3.4, 3.9], [7.0, 7.4]]

Frames 6–6 · 2.40–2.90s (end exclusive)

Code indices f5–f5 (0-based, 2 FPS sampled sequence); source files 00001373.jpg → 00001373.jpg. Last observed frame time: 2.40s.

Interval boundary estimated by the model from saved frames, not human ground truth.

Frames 8–8 · 3.40–3.90s (end exclusive)

Code indices f7–f7 (0-based, 2 FPS sampled sequence); source files 00001388.jpg → 00001388.jpg. Last observed frame time: 3.40s.

Interval boundary estimated by the model from saved frames, not human ground truth.

Frames 15–15 · 7.00–7.40s (end exclusive)

Code indices f14–f14 (0-based, 2 FPS sampled sequence); source files 00001442.jpg → 00001442.jpg. Last observed frame time: 7.00s.

Interval boundary estimated by the model from saved frames, not human ground truth.

glidessimple · highSubject: A large fish g1 → #29

Located · Localization confidence high: The large foreground fish passes along the upper scene and becomes prominently visible close by.

Frame sequence:[[12, 18]]Time sequence:[[6.0, 9.2]]

Frames 13–19 · 6.00–9.20s (end exclusive)

Code indices f12–f18 (0-based, 2 FPS sampled sequence); source files 00001427.jpg → 00001472.jpg. Last observed frame time: 9.00s.

glides first action frame — First observed frame · Frame 13

glides last action frame — Last observed frame · Frame 19

Clip end reached; the true end may be later

g0 → #3 · person
A girl / She Category check: agreeg1 → #29 · other animal
many long-snouted fish / them / A large fish Category check: agreeg2 → #18 · other animal
many long-snouted fish / them Category check: agreeg3 → #10 · other animal
many long-snouted fish / them Category check: agreeg4 → #7 · other animal
many long-snouted fish / them Category check: agreeg5 → #12 · other animal
many long-snouted fish / them Category check: agreeg6 → #6 · other animal
many long-snouted fish / them Category check: agreeg7 → #8 · other animal
many long-snouted fish / them Category check: agreeg8 → #11 · other animal
many long-snouted fish / them Category check: agreeg9 → #5 · other animal
many long-snouted fish / them Category check: agreeg10 → #4 · other animal
many long-snouted fish / them Category check: agreeg11 → #16 · other animal
many long-snouted fish / them Category check: agreeg12 → #19 · other animal
many long-snouted fish / them Category check: agreeg13 → #15 · other animal
many long-snouted fish / them Category check: agreeg14 → #22 · other animal
many long-snouted fish / them Category check: agreeg15 → #14 · other animal
many long-snouted fish / them Category check: agreeg16 → #21 · other animal
many long-snouted fish / them Category check: agreeg17 → #17 · other animal
many long-snouted fish / them Category check: agreeg18 → #13 · other animal
many long-snouted fish / them Category check: agreeg19 → #25 · other animal
many long-snouted fish / them Category check: agreeg20 → #24 · other animal
many long-snouted fish / them Category check: agreeg21 → #23 · other animal
many long-snouted fish / them Category check: agreeg22 → #31 · other animal
many long-snouted fish / them Category check: agreeg23 → #9 · other animal
many long-snouted fish / them Category check: agreeg24 → #26 · other animal
many long-snouted fish / them Category check: agreeg25 → #30 · other animal
many long-snouted fish / them Category check: agreeg26 → #27 · other animal
many long-snouted fish / them Category check: agreeg27 → #28 · other animal
many long-snouted fish / them Category check: agreeg28 → #32 · other animal
many long-snouted fish / them Category check: agreeg29 → #20 · other animal
many long-snouted fish / them Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 30/32; subject/object links: 6; unlinked: 0.

390_TDIMeOVXQr0

5 actionsText check: ok9.2s · 1 model call

Frame 10 (f9) · source frame 668 · 4.40s

Frame 11 (f10) · source frame 677 · 5.00s

Frame 12 (f11) · source frame 683 · 5.40s

Frame 13 (f12) · source frame 692 · 6.00s

Frame 14 (f13) · source frame 698 · 6.40s

Frame 15 (f14) · source frame 707 · 7.00s

Frame 16 (f15) · source frame 713 · 7.40s

Frame 17 (f16) · source frame 722 · 8.00s

Frame 18 (f17) · source frame 728 · 8.40s

Frame 19 (f18) · source frame 737 · 9.00s

Caption + Actions + Temporal Localization

Many people are beside black lockers. Some walk past them, and others sit. One seated boy stands up and turns toward the lockers.

walksimple · highSubject: Some g2 → #18, g6 → #8, g7 → #15, g8 → #9, g9 → #5, g10 → #7, g12 → #23, g13 → #12Object: them g23 → #3

Located · Localization confidence medium: Different members walk past the lockers early, across the foreground midway, and through the room near the end.

Frame sequence:[[0, 5], [6, 10], [14, 18]]Time sequence:[[0.0, 2.9], [3.0, 5.4], [7.0, 9.2]]

Frames 1–6 · 0.00–2.90s (end exclusive)

Code indices f0–f5 (0-based, 2 FPS sampled sequence); source files 00000602.jpg → 00000638.jpg. Last observed frame time: 2.40s.

walk first action frame — First observed frame · Frame 1

walk last action frame — Last observed frame · Frame 6

Clip start reached; the true start may be earlier

Frames 7–11 · 3.00–5.40s (end exclusive)

Code indices f6–f10 (0-based, 2 FPS sampled sequence); source files 00000647.jpg → 00000677.jpg. Last observed frame time: 5.00s.

Interval boundary estimated by the model from saved frames, not human ground truth.

Frames 15–19 · 7.00–9.20s (end exclusive)

Code indices f14–f18 (0-based, 2 FPS sampled sequence); source files 00000707.jpg → 00000737.jpg. Last observed frame time: 9.00s.

Clip end reached; the true end may be later

sitposture · highSubject: others g0 → #11, g1 → #10, g4 → #16

Located · Localization confidence high: People remain seated along the lockers; the number-nine boy leaves his seat before the other seated people disappear from view.

Frame sequence:[[0, 13]]Time sequence:[[0.0, 6.9]]

Frames 1–14 · 0.00–6.90s (end exclusive)

Code indices f0–f13 (0-based, 2 FPS sampled sequence); source files 00000602.jpg → 00000698.jpg. Last observed frame time: 6.40s.

sit first action frame — First observed frame · Frame 1

sit last action frame — Last observed frame · Frame 14

Clip start reached; the true start may be earlier

seatedposture · highSubject: One seated boy g1 → #10

Located · Localization confidence high: The number-nine boy sits at the lockers before beginning to rise.

Frame sequence:[[0, 10]]Time sequence:[[0.0, 5.4]]

Frames 1–11 · 0.00–5.40s (end exclusive)

Code indices f0–f10 (0-based, 2 FPS sampled sequence); source files 00000602.jpg → 00000677.jpg. Last observed frame time: 5.00s.

seated first action frame — First observed frame · Frame 1

seated last action frame — Last observed frame · Frame 11

Clip start reached; the true start may be earlier

stands upphrasal · highSubject: One seated boy g1 → #10

Located · Localization confidence high: The boy lifts himself from his seat and straightens his legs.

Frame sequence:[[10, 11]]Time sequence:[[5.0, 5.9]]

Frames 11–12 · 5.00–5.90s (end exclusive)

Code indices f10–f11 (0-based, 2 FPS sampled sequence); source files 00000677.jpg → 00000683.jpg. Last observed frame time: 5.40s.

stands up first action frame — First observed frame · Frame 11

stands up last action frame — Last observed frame · Frame 12

Interval boundary estimated by the model from saved frames, not human ground truth.

turnssimple · highSubject: One seated boy g1 → #10Object: the lockers g23 → #3

Located · Localization confidence high: After rising, the boy rotates until his back faces the viewer and he faces the lockers.

Frame sequence:[[11, 12]]Time sequence:[[5.4, 6.4]]

Frames 12–13 · 5.40–6.40s (end exclusive)

Code indices f11–f12 (0-based, 2 FPS sampled sequence); source files 00000683.jpg → 00000692.jpg. Last observed frame time: 6.00s.

turns first action frame — First observed frame · Frame 12

turns last action frame — Last observed frame · Frame 13

Interval boundary estimated by the model from saved frames, not human ground truth.

g0 → #11 · person
Many people / others Category check: agreeg1 → #10 · person
Many people / others / One seated boy Category check: agreeg2 → #18 · person
Many people / Some Category check: agreeg3 → #14 · person
Many people Category check: agreeg4 → #16 · person
Many people / others Category check: agreeg5 → #21 · person
Many people Category check: agreeg6 → #8 · person
Many people / Some Category check: agreeg7 → #15 · person
Many people / Some Category check: agreeg8 → #9 · person
Many people / Some Category check: agreeg9 → #5 · person
Many people / Some Category check: agreeg10 → #7 · person
Many people / Some Category check: agreeg11 → #6 · person
Many people Category check: agreeg12 → #23 · person
Many people / Some Category check: agreeg13 → #12 · person
Many people / Some Category check: agreeg14 → #22 · person
Many people Category check: agreeg15 → #17 · person
Many people Category check: agreeg16 → #24 · person
Many people Category check: agreeg17 → #27 · person
Many people Category check: agreeg18 → #13 · person
Many people Category check: agreeg19 → #26 · person
Many people Category check: agreeg20 → #29 · person
Many people Category check: agreeg21 → #28 · person
Many people Category check: agreeg22 → #30 · person
Many people Category check: agreeg23 → #3 · cupboard or showcase or storage rack
black lockers / them / the lockers Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 24/30; subject/object links: 7; unlinked: 0.

473_PWeWs8Uf9TA

1 actionsText check: ok9.2s · 1 model call

Frame 10 (f9) · source frame 218 · 4.40s

Frame 11 (f10) · source frame 227 · 5.00s

Frame 12 (f11) · source frame 233 · 5.40s

Frame 13 (f12) · source frame 242 · 6.00s

Frame 14 (f13) · source frame 248 · 6.40s

Frame 15 (f14) · source frame 257 · 7.00s

Frame 16 (f15) · source frame 263 · 7.40s

Frame 17 (f16) · source frame 272 · 8.00s

Frame 18 (f17) · source frame 278 · 8.40s

Frame 19 (f18) · source frame 287 · 9.00s

Caption + Actions + Temporal Localization

A person stands inside a transparent dome beside a wooden deck. Two ladders, two tables, plant pots and bags occupy the deck. A backhoe and a cement mixer are near a smaller dome.

standsposture · highSubject: A person g0 → #10Object: a transparent dome g1 → #3

Located · Localization confidence high: The person is upright inside the doorway in the first three clean frames before becoming obscured.

Frame sequence:[[0, 2]]Time sequence:[[0.0, 1.4]]

Frames 1–3 · 0.00–1.40s (end exclusive)

Code indices f0–f2 (0-based, 2 FPS sampled sequence); source files 00000152.jpg → 00000167.jpg. Last observed frame time: 1.00s.

Clip start reached; the true start may be earlier

g0 → #10 · person
A person Category check: agreeg1 → #3 · other construction
a transparent dome / a smaller dome Category check: unknowng2 → #1 · ground
a wooden deck / the deck Category check: unknowng3 → #8 · ladder
Two ladders Category check: agreeg4 → #9 · ladder
Two ladders Category check: agreeg5 → #12 · table or desk
two tables Category check: agreeg6 → #13 · table or desk
two tables Category check: agreeg7 → #14 · flower pot or vase
plant pots Category check: agreeg8 → #15 · flower pot or vase
plant pots Category check: agreeg9 → #19 · flower pot or vase
plant pots Category check: agreeg10 → #11 · bag or package
bags Category check: agreeg11 → #17 · bag or package
bags Category check: agreeg12 → #18 · bag or package
bags Category check: agreeg13 → #16 · wheeled machine
A backhoe / a cement mixer Category check: disagree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 14/20; subject/object links: 2; unlinked: 0.

562_OA61thiz9wU

1 actionsText check: ok8.2s · 1 model call

Caption + Actions + Temporal Localization

A gray naval ship moves slowly right along the river. It is visible beyond a metal fence near a large leafy tree.

movessimple · highSubject: A gray naval ship g0 → #7

Located · Localization confidence high: The ship steadily advances right relative to the fence and tree throughout the sequence.

Frame sequence:[[0, 16]]Time sequence:[[0.0, 8.2]]

Frames 1–17 · 0.00–8.20s (end exclusive)

Code indices f0–f16 (0-based, 2 FPS sampled sequence); source files 00000152.jpg → 00000272.jpg. Last observed frame time: 8.00s.

moves first action frame — First observed frame · Frame 1

moves last action frame — Last observed frame · Frame 17

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #7 · ship or boat
A gray naval ship / It Category check: agreeg1 → #5 · river
the river Category check: agreeg2 → #1 · handrail or fence
a metal fence Category check: agreeg3 → #6 · tree
a large leafy tree Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 4/7; subject/object links: 1; unlinked: 0.

64_8_IdHgf2G6g

5 actionsText check: ok7.2s · 1 model call

Frame 10 (f9) · source frame 758 · 4.40s

Frame 11 (f10) · source frame 767 · 5.00s

Frame 12 (f11) · source frame 773 · 5.40s

Frame 13 (f12) · source frame 782 · 6.00s

Frame 14 (f13) · source frame 788 · 6.40s

Frame 15 (f14) · source frame 797 · 7.00s

Caption + Actions + Temporal Localization

A yellow-shirted player skates across the ice. A white-shirted player makes tight turns holding a hockey stick. This player then skates toward a goalie standing before the net.

skatessimple · highSubject: A yellow-shirted player g0 → #10

Located · Localization confidence high: The yellow-shirted player glides rightward with alternating skating strides.

Frame sequence:[[0, 3]]Time sequence:[[0.0, 1.9]]

Frames 1–4 · 0.00–1.90s (end exclusive)

Code indices f0–f3 (0-based, 2 FPS sampled sequence); source files 00000692.jpg → 00000713.jpg. Last observed frame time: 1.40s.

skates first action frame — First observed frame · Frame 1

skates last action frame — Last observed frame · Frame 4

Clip start reached; the true start may be earlier

makes tight turnslight_verb · highSubject: A white-shirted player g2 → #14

Located · Localization confidence high: The white-shirted player leans into successive curves and changes skating direction.

Frame sequence:[[7, 12]]Time sequence:[[3.4, 6.4]]

Frames 8–13 · 3.40–6.40s (end exclusive)

Code indices f7–f12 (0-based, 2 FPS sampled sequence); source files 00000743.jpg → 00000782.jpg. Last observed frame time: 6.00s.

makes tight turns first action frame — First observed frame · Frame 8

makes tight turns last action frame — Last observed frame · Frame 13

Interval boundary estimated by the model from saved frames, not human ground truth.

holdingsimple · highSubject: A white-shirted player g2 → #14Object: a hockey stick g3 → #5

Located · Localization confidence high: The player visibly grips the hockey stick throughout the skating sequence.

Frame sequence:[[4, 14]]Time sequence:[[2.0, 7.2]]

Frames 5–15 · 2.00–7.20s (end exclusive)

Code indices f4–f14 (0-based, 2 FPS sampled sequence); source files 00000722.jpg → 00000797.jpg. Last observed frame time: 7.00s.

holding first action frame — First observed frame · Frame 5

holding last action frame — Last observed frame · Frame 15

Clip end reached; the true end may be later

skatessimple · highSubject: This player g2 → #14Object: a goalie g4 → #12

Located · Localization confidence high: After the turns, the player advances toward the goalie and net.

Frame sequence:[[12, 14]]Time sequence:[[6.0, 7.2]]

Frames 13–15 · 6.00–7.20s (end exclusive)

Code indices f12–f14 (0-based, 2 FPS sampled sequence); source files 00000782.jpg → 00000797.jpg. Last observed frame time: 7.00s.

Clip end reached; the true end may be later

standingposture · highSubject: a goalie g4 → #12Object: the net g5 → #7

Located · Localization confidence high: The goalie remains upright on skates in front of the net.

Frame sequence:[[10, 14]]Time sequence:[[5.0, 7.2]]

Frames 11–15 · 5.00–7.20s (end exclusive)

Code indices f10–f14 (0-based, 2 FPS sampled sequence); source files 00000767.jpg → 00000797.jpg. Last observed frame time: 7.00s.

standing first action frame — First observed frame · Frame 11

standing last action frame — Last observed frame · Frame 15

Clip end reached; the true end may be later

g0 → #10 · person
A yellow-shirted player Category check: agreeg1 → #3 · ice
the ice Category check: agreeg2 → #14 · person
A white-shirted player / This player Category check: agreeg3 → #5 · bat
a hockey stick Category check: disagreeg4 → #12 · person
a goalie Category check: unknowng5 → #7 · goal
the net Category check: disagree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 6/28; subject/object links: 8; unlinked: 0.

737_GU3ZbsUq3MA

4 actionsText check: ok9.2s · 1 model call

Caption + Actions + Temporal Localization

A masked man holds a brush beside rows of blue seats. He reaches toward the right-hand bags, then gestures. Behind him, an orange-shirted man moves a long tool along the left window.

holdssimple · highSubject: A masked man g0 → #8Object: a brush g1 → #4

Located · Localization confidence high: The masked man holds the upright brush throughout the observed sequence.

Frame sequence:[[0, 18]]Time sequence:[[0.0, 9.2]]

Frames 1–19 · 0.00–9.20s (end exclusive)

Code indices f0–f18 (0-based, 2 FPS sampled sequence); source files 00000602.jpg → 00000737.jpg. Last observed frame time: 9.00s.

holds first action frame — First observed frame · Frame 1

holds last action frame — Last observed frame · Frame 19

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

reachesprepositional · highSubject: He g0 → #8Object: the right-hand bags g14 → #9, g15 → #23

Located · Localization confidence high: He extends toward the bags on his right, retracts, and then reaches down toward them again.

Frame sequence:[[4, 5], [7, 9]]Time sequence:[[2.0, 2.9], [3.4, 4.9]]

Frames 5–6 · 2.00–2.90s (end exclusive)

Code indices f4–f5 (0-based, 2 FPS sampled sequence); source files 00000632.jpg → 00000638.jpg. Last observed frame time: 2.40s.

Interval boundary estimated by the model from saved frames, not human ground truth.

Frames 8–10 · 3.40–4.90s (end exclusive)

Code indices f7–f9 (0-based, 2 FPS sampled sequence); source files 00000653.jpg → 00000668.jpg. Last observed frame time: 4.40s.

Interval boundary estimated by the model from saved frames, not human ground truth.

gesturessimple · highSubject: He g0 → #8

Located · Localization confidence high: He repeatedly raises and changes the position of his free gloved hand.

Frame sequence:[[10, 18]]Time sequence:[[5.0, 9.2]]

Frames 11–19 · 5.00–9.20s (end exclusive)

Code indices f10–f18 (0-based, 2 FPS sampled sequence); source files 00000677.jpg → 00000737.jpg. Last observed frame time: 9.00s.

Clip end reached; the true end may be later

movessimple · mediumSubject: an orange-shirted man g16 → #17Object: a long tool g1 → #4Object: the left window g17 → #19

Located · Localization confidence medium: The orange-shirted man shifts the long tool along the left window area before lifting it away.

Frame sequence:[[4, 15]]Time sequence:[[2.0, 7.9]]

Frames 5–16 · 2.00–7.90s (end exclusive)

Code indices f4–f15 (0-based, 2 FPS sampled sequence); source files 00000632.jpg → 00000713.jpg. Last observed frame time: 7.40s.

Interval boundary estimated by the model from saved frames, not human ground truth.

g0 → #8 · person
A masked man / He / him Category check: agreeg1 → #4 · tool
a brush / a long tool Category check: agreeg2 → #10 · chair or seat
rows of blue seats Category check: agreeg3 → #14 · chair or seat
rows of blue seats Category check: agreeg4 → #11 · chair or seat
rows of blue seats Category check: agreeg5 → #21 · chair or seat
rows of blue seats Category check: agreeg6 → #18 · chair or seat
rows of blue seats Category check: agreeg7 → #15 · chair or seat
rows of blue seats Category check: agreeg8 → #22 · chair or seat
rows of blue seats Category check: agreeg9 → #12 · chair or seat
rows of blue seats Category check: agreeg10 → #13 · chair or seat
rows of blue seats Category check: agreeg11 → #24 · chair or seat
rows of blue seats Category check: agreeg12 → #16 · chair or seat
rows of blue seats Category check: agreeg13 → #25 · chair or seat
rows of blue seats Category check: agreeg14 → #9 · bag or package
the right-hand bags Category check: agreeg15 → #23 · bag or package
the right-hand bags Category check: agreeg16 → #17 · person
an orange-shirted man Category check: agreeg17 → #19 · window
the left window Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 18/25; subject/object links: 8; unlinked: 0.

826_0bniEAy5YfI

3 actionsText check: ok5.2s · 1 model call

Frame 10 (f9) · source frame 143 · 4.40s

Frame 11 (f10) · source frame 152 · 5.00s

Caption + Actions + Temporal Localization

Four players play badminton across a net indoors. They move sideways on the wooden floor during the rally. A man sits on a raised chair.

play badmintonnonvisual_object · highSubject: Four players g0 → #7, g1 → #8, g2 → #9, g3 → #11Object: a net g4 → #5

Located · Localization confidence high: The players exchange shots and react to the rally after initially holding ready positions.

Frame sequence:[[4, 10]]Time sequence:[[2.0, 5.2]]

Frames 5–11 · 2.00–5.20s (end exclusive)

Code indices f4–f10 (0-based, 2 FPS sampled sequence); source files 00000107.jpg → 00000152.jpg. Last observed frame time: 5.00s.

play badminton first action frame — First observed frame · Frame 5

play badminton last action frame — Last observed frame · Frame 11

Clip end reached; the true end may be later

movesimple · highSubject: They g0 → #7, g1 → #8, g2 → #9, g3 → #11

Located · Localization confidence medium: Both pairs reposition laterally during the rally, although their individual movements occur at different moments.

Frame sequence:[[4, 10]]Time sequence:[[2.0, 5.2]]

Frames 5–11 · 2.00–5.20s (end exclusive)

Code indices f4–f10 (0-based, 2 FPS sampled sequence); source files 00000107.jpg → 00000152.jpg. Last observed frame time: 5.00s.

move first action frame — First observed frame · Frame 5

move last action frame — Last observed frame · Frame 11

Clip end reached; the true end may be later

sitsposture · highSubject: A man g6 → #10Object: a raised chair g7 → #6

Located · Localization confidence high: The man remains seated on the elevated chair throughout the sequence.

Frame sequence:[[0, 10]]Time sequence:[[0.0, 5.2]]

Frames 1–11 · 0.00–5.20s (end exclusive)

Code indices f0–f10 (0-based, 2 FPS sampled sequence); source files 00000077.jpg → 00000152.jpg. Last observed frame time: 5.00s.

sits first action frame — First observed frame · Frame 1

sits last action frame — Last observed frame · Frame 11

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #7 · person
Four players / They Category check: agreeg1 → #8 · person
Four players / They Category check: agreeg2 → #9 · person
Four players / They Category check: agreeg3 → #11 · person
Four players / They Category check: agreeg4 → #5 · ball net
a net Category check: agreeg5 → #3 · floor
the wooden floor Category check: agreeg6 → #10 · person
A man Category check: agreeg7 → #6 · shelf
a raised chair Category check: disagree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 8/16; subject/object links: 5; unlinked: 0.

912_bGghFu3zO-c

2 actionsText check: ok3.0s · 1 model call

Frame 1 (f0) · source frame 3896 · 0.00s

Frame 2 (f1) · source frame 3902 · 0.40s

Frame 3 (f2) · source frame 3911 · 1.00s

Frame 4 (f3) · source frame 3917 · 1.40s

Frame 5 (f4) · source frame 3926 · 2.00s

Frame 6 (f5) · source frame 3932 · 2.40s

Frame 7 (f6) · source frame 3938 · 2.80s

Caption + Actions + Temporal Localization

Four people sit in a red airplane-shaped ride outdoors. It moves right along an elevated track.

sitposture · highSubject: Four people g0 → #10, g1 → #8, g2 → #11, g3 → #9Object: a red airplane-shaped ride g4 → #6

Located · Localization confidence high: All four riders remain seated throughout the sequence, although the rear riders are partly obscured.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00003896.jpg → 00003938.jpg. Last observed frame time: 2.80s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

movessimple · highSubject: It g4 → #6Object: an elevated track g5 → #3

Located · Localization confidence high: The ride advances rightward along the track relative to the fixed supports and fence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00003896.jpg → 00003938.jpg. Last observed frame time: 2.80s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #10 · person
Four people Category check: agreeg1 → #8 · person
Four people Category check: agreeg2 → #11 · person
Four people Category check: agreeg3 → #9 · person
Four people Category check: agreeg4 → #6 · toy
a red airplane-shaped ride / It Category check: unknowng5 → #3 · other construction
an elevated track Category check: disagree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 6/11; subject/object links: 4; unlinked: 0.

9_E0zfiF4DCt8

3 actionsText check: ok9.2s · 1 model call

Caption + Actions + Temporal Localization

A red-hooded skier gestures on a snowy slope. They remain standing. Three other skiers glide downhill.

gesturessimple · highSubject: A red-hooded skier g0 → #6

Located · Localization confidence medium: The red-hooded skier moves their gloved hands from near their face to forward positions.

Frame sequence:[[0, 5]]Time sequence:[[0.0, 2.9]]

Frames 1–6 · 0.00–2.90s (end exclusive)

Code indices f0–f5 (0-based, 2 FPS sampled sequence); source files 00000152.jpg → 00000188.jpg. Last observed frame time: 2.40s.

Clip start reached; the true start may be earlier

standingposture · highSubject: They g0 → #6

Located · Localization confidence high: The red-hooded skier maintains an upright, slightly bent standing posture throughout.

Frame sequence:[[0, 18]]Time sequence:[[0.0, 9.2]]

Frames 1–19 · 0.00–9.20s (end exclusive)

Code indices f0–f18 (0-based, 2 FPS sampled sequence); source files 00000152.jpg → 00000287.jpg. Last observed frame time: 9.00s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

glidesimple · highSubject: Three other skiers g2 → #7, g3 → #8, g4 → #9

Located · Localization confidence medium: The skiers descend successively: the first is visible throughout, the second enters around frame 6, and the third appears around frame 14.

Frame sequence:[[0, 18]]Time sequence:[[0.0, 9.2]]

Frames 1–19 · 0.00–9.20s (end exclusive)

Code indices f0–f18 (0-based, 2 FPS sampled sequence); source files 00000152.jpg → 00000287.jpg. Last observed frame time: 9.00s.

glide first action frame — First observed frame · Frame 1

glide last action frame — Last observed frame · Frame 19

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #6 · person
A red-hooded skier / They Category check: unknowng1 → #4 · snowfield
a snowy slope Category check: disagreeg2 → #7 · person
Three other skiers Category check: unknowng3 → #8 · person
Three other skiers Category check: unknowng4 → #9 · person
Three other skiers Category check: unknown

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 5/9; subject/object links: 3; unlinked: 0.

1039_QZROp4t0doU

1 actionsText check: ok9.2s · 1 model call

Frame 11 (f10) · source frame 92 · 5.00s

Frame 12 (f11) · source frame 98 · 5.40s

Frame 13 (f12) · source frame 107 · 6.00s

Frame 14 (f13) · source frame 113 · 6.40s

Frame 15 (f14) · source frame 122 · 7.00s

Frame 16 (f15) · source frame 128 · 7.40s

Frame 17 (f16) · source frame 137 · 8.00s

Frame 18 (f17) · source frame 143 · 8.40s

Frame 19 (f18) · source frame 152 · 9.00s

Caption + Actions + Temporal Localization

A motorcycle travels along a straight road between green fields and trees. It heads away into the distance.

travels / headssimple · highSubject: A motorcycle g0 → #4

Located · Localization confidence high: The motorcycle appears near the bottom of the road and progresses toward its distant end.

Frame sequence:[[3, 18]]Time sequence:[[1.4, 9.2]]

Frames 4–19 · 1.40–9.20s (end exclusive)

Code indices f3–f18 (0-based, 2 FPS sampled sequence); source files 00000038.jpg → 00000152.jpg. Last observed frame time: 9.00s.

Clip end reached; the true end may be later

g0 → #4 · motorcycle
A motorcycle / It Category check: agreeg1 → #2 · road
a straight road Category check: agreeg2 → #1 · grass
green fields Category check: agreeg3 → #3 · tree
trees Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 4/6; subject/object links: 1; unlinked: 0.

1129_eM5eaNKgIHo

2 actionsText check: ok3.0s · 1 model call

Caption + Actions + Temporal Localization

A turkey walks across the grass beside gravestones and a small American flag. It looks left.

walkssimple · highSubject: A turkey g0 → #5Object: gravestones g2 → #2Object: a small American flag g3 → #4

Located · Localization confidence high: The turkey takes successive steps across the grass throughout the sequence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00000243.jpg → 00000285.jpg. Last observed frame time: 2.80s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

lookssimple · mediumSubject: It g0 → #5

Located · Localization confidence medium: The turkey visibly faces left in the final two frames.

Frame sequence:[[5, 6]]Time sequence:[[2.4, 3.0]]

Frames 6–7 · 2.40–3.00s (end exclusive)

Code indices f5–f6 (0-based, 2 FPS sampled sequence); source files 00000279.jpg → 00000285.jpg. Last observed frame time: 2.80s.

Clip end reached; the true end may be later

g0 → #5 · other animal
A turkey / It Category check: unknowng1 → #1 · grass
the grass Category check: unknowng2 → #2 · stone
gravestones Category check: unknowng3 → #4 · flag
a small American flag Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 4/6; subject/object links: 4; unlinked: 0.

1215_NSw8hNtpSy0

1 actionsText check: ok1.6s · 1 model call

Frame 1 (f0) · source frame 6031 · 0.00s

Frame 2 (f1) · source frame 6037 · 0.40s

Frame 3 (f2) · source frame 6046 · 1.00s

Frame 4 (f3) · source frame 6052 · 1.40s

Caption + Actions + Temporal Localization

An orange sports car is on a glossy floor indoors. Two people in red stand behind it near a wall-mounted television.

standposture · highSubject: Two people g2 → #8, g3 → #7Object: it g0 → #6Object: a wall-mounted television g4 → #9

Located · Localization confidence medium: Both red-clad people remain upright behind the car throughout the sequence, although their lower bodies are obscured.

Frame sequence:[[0, 3]]Time sequence:[[0.0, 1.6]]

Frames 1–4 · 0.00–1.60s (end exclusive)

Code indices f0–f3 (0-based, 2 FPS sampled sequence); source files 00006031.jpg → 00006052.jpg. Last observed frame time: 1.40s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #6 · car
An orange sports car / it Category check: agreeg1 → #4 · floor
a glossy floor Category check: agreeg2 → #8 · person
Two people Category check: agreeg3 → #7 · person
Two people Category check: agreeg4 → #9 · screen or television
a wall-mounted television Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 5/9; subject/object links: 3; unlinked: 0.

12_1mWNahzcsAc

4 actionsText check: ok3.0s · 1 model call

Frame 1 (f0) · source frame 1255 · 0.00s

Frame 2 (f1) · source frame 1261 · 0.40s

Frame 3 (f2) · source frame 1270 · 1.00s

Frame 4 (f3) · source frame 1276 · 1.40s

Frame 5 (f4) · source frame 1285 · 2.00s

Frame 6 (f5) · source frame 1291 · 2.40s

Frame 7 (f6) · source frame 1297 · 2.80s

Caption + Actions + Temporal Localization

A woman in pink strokes a dark horse. It bends down beside a woman in black. She holds a pale horse as it turns right.

strokessimple · highSubject: A woman g0 → #7Object: a dark horse g1 → #10

Located · Localization confidence high: The woman moves her hand down the dark horse's face as it lowers.

Frame sequence:[[0, 3]]Time sequence:[[0.0, 1.9]]

Frames 1–4 · 0.00–1.90s (end exclusive)

Code indices f0–f3 (0-based, 2 FPS sampled sequence); source files 00001255.jpg → 00001276.jpg. Last observed frame time: 1.40s.

strokes first action frame — First observed frame · Frame 1

strokes last action frame — Last observed frame · Frame 4

Clip start reached; the true start may be earlier

bendssimple · highSubject: It g1 → #10Object: a woman g2 → #8

Located · Localization confidence medium: The dark horse bends its neck downward, briefly raises it, then bends farther down near the end.

Frame sequence:[[0, 3], [5, 6]]Time sequence:[[0.0, 1.9], [2.4, 3.0]]

Frames 1–4 · 0.00–1.90s (end exclusive)

Code indices f0–f3 (0-based, 2 FPS sampled sequence); source files 00001255.jpg → 00001276.jpg. Last observed frame time: 1.40s.

bends first action frame — First observed frame · Frame 1

bends last action frame — Last observed frame · Frame 4

Clip start reached; the true start may be earlier

Frames 6–7 · 2.40–3.00s (end exclusive)

Code indices f5–f6 (0-based, 2 FPS sampled sequence); source files 00001291.jpg → 00001297.jpg. Last observed frame time: 2.80s.

Clip end reached; the true end may be later

holdssimple · highSubject: She g2 → #8Object: a pale horse g3 → #9

Located · Localization confidence high: The woman maintains her grip near the pale horse's halter throughout the sequence.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00001255.jpg → 00001297.jpg. Last observed frame time: 2.80s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

turnssimple · highSubject: it g3 → #9

Located · Localization confidence high: The pale horse progressively turns its head toward the right side of the image.

Frame sequence:[[3, 6]]Time sequence:[[1.4, 3.0]]

Frames 4–7 · 1.40–3.00s (end exclusive)

Code indices f3–f6 (0-based, 2 FPS sampled sequence); source files 00001276.jpg → 00001297.jpg. Last observed frame time: 2.80s.

Clip end reached; the true end may be later

g0 → #7 · person
A woman Category check: agreeg1 → #10 · horse
a dark horse / It Category check: agreeg2 → #8 · person
a woman / She Category check: agreeg3 → #9 · horse
a pale horse / it Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 4/10; subject/object links: 7; unlinked: 0.

1409_44et30baPhs

3 actionsText check: ok3.0s · 1 model call

Frame 1 (f0) · source frame 2208 · 0.00s

Frame 2 (f1) · source frame 2214 · 0.40s

Frame 3 (f2) · source frame 2223 · 1.00s

Frame 4 (f3) · source frame 2229 · 1.40s

Frame 5 (f4) · source frame 2238 · 2.00s

Frame 6 (f5) · source frame 2244 · 2.40s

Frame 7 (f6) · source frame 2250 · 2.80s

Caption + Actions + Temporal Localization

A man points toward books on a shelf. He turns forward. He then gestures toward the books.

pointsprepositional · mediumSubject: A man g0 → #6Object: books g1 → #5

Located · Localization confidence medium: His extended hand indicates the books, at times making contact with them.

Frame sequence:[[0, 4]]Time sequence:[[0.0, 2.4]]

Frames 1–5 · 0.00–2.40s (end exclusive)

Code indices f0–f4 (0-based, 2 FPS sampled sequence); source files 00002208.jpg → 00002238.jpg. Last observed frame time: 2.00s.

points first action frame — First observed frame · Frame 1

points last action frame — Last observed frame · Frame 5

Clip start reached; the true start may be earlier

turnssimple · highSubject: He g0 → #6

Located · Localization confidence high: He changes from facing the shelves to facing forward.

Frame sequence:[[0, 2]]Time sequence:[[0.0, 1.4]]

Frames 1–3 · 0.00–1.40s (end exclusive)

Code indices f0–f2 (0-based, 2 FPS sampled sequence); source files 00002208.jpg → 00002223.jpg. Last observed frame time: 1.00s.

Clip start reached; the true start may be earlier

gesturesprepositional · highSubject: He g0 → #6Object: the books g1 → #5

Located · Localization confidence high: He opens his hand and extends it toward the books.

Frame sequence:[[5, 6]]Time sequence:[[2.4, 3.0]]

Frames 6–7 · 2.40–3.00s (end exclusive)

Code indices f5–f6 (0-based, 2 FPS sampled sequence); source files 00002244.jpg → 00002250.jpg. Last observed frame time: 2.80s.

Clip end reached; the true end may be later

g0 → #6 · person
A man / He Category check: agreeg1 → #5 · book
books / the books Category check: agreeg2 → #4 · cupboard or showcase or storage rack
a shelf Category check: disagree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 3/7; subject/object links: 5; unlinked: 0.

155_GVuOmIPfHj0

2 actionsText check: ok9.2s · 1 model call

Frame 1 (f0) · source frame 1772 · 0.00s

Frame 2 (f1) · source frame 1778 · 0.40s

Frame 3 (f2) · source frame 1787 · 1.00s

Frame 4 (f3) · source frame 1793 · 1.40s

Frame 5 (f4) · source frame 1802 · 2.00s

Frame 6 (f5) · source frame 1808 · 2.40s

Frame 7 (f6) · source frame 1817 · 3.00s

Frame 8 (f7) · source frame 1823 · 3.40s

Frame 9 (f8) · source frame 1832 · 4.00s

Frame 10 (f9) · source frame 1838 · 4.40s

Frame 11 (f10) · source frame 1847 · 5.00s

Frame 12 (f11) · source frame 1853 · 5.40s

Frame 13 (f12) · source frame 1862 · 6.00s

Frame 14 (f13) · source frame 1868 · 6.40s

Frame 15 (f14) · source frame 1877 · 7.00s

Frame 16 (f15) · source frame 1883 · 7.40s

Frame 17 (f16) · source frame 1892 · 8.00s

Frame 18 (f17) · source frame 1898 · 8.40s

Frame 19 (f18) · source frame 1907 · 9.00s

Caption + Actions + Temporal Localization

An orange excavator raises its arm behind a parked truck. It then rotates to the right on the dirt ground.

raisessimple · highSubject: An orange excavator g0 → #4Object: its arm g0 → #4Object: a parked truck g1 → #6

Located · Localization confidence medium: The articulated arm rises initially and makes further upward movements after visible pauses.

Frame sequence:[[0, 2], [8, 10], [14, 16]]Time sequence:[[0.0, 1.4], [4.0, 5.4], [7.0, 8.4]]

Frames 1–3 · 0.00–1.40s (end exclusive)

Code indices f0–f2 (0-based, 2 FPS sampled sequence); source files 00001772.jpg → 00001787.jpg. Last observed frame time: 1.00s.

raises first action frame — First observed frame · Frame 1

raises last action frame — Last observed frame · Frame 3

Clip start reached; the true start may be earlier

Frames 9–11 · 4.00–5.40s (end exclusive)

Code indices f8–f10 (0-based, 2 FPS sampled sequence); source files 00001832.jpg → 00001847.jpg. Last observed frame time: 5.00s.

Interval boundary estimated by the model from saved frames, not human ground truth.

Frames 15–17 · 7.00–8.40s (end exclusive)

Code indices f14–f16 (0-based, 2 FPS sampled sequence); source files 00001877.jpg → 00001892.jpg. Last observed frame time: 8.00s.

Interval boundary estimated by the model from saved frames, not human ground truth.

rotatessimple · highSubject: It g0 → #4

Located · Localization confidence medium: The excavator's upper body turns right, pauses, and turns farther right near the end.

Frame sequence:[[4, 6], [16, 18]]Time sequence:[[2.0, 3.4], [8.0, 9.2]]

Frames 5–7 · 2.00–3.40s (end exclusive)

Code indices f4–f6 (0-based, 2 FPS sampled sequence); source files 00001802.jpg → 00001817.jpg. Last observed frame time: 3.00s.

rotates first action frame — First observed frame · Frame 5

rotates last action frame — Last observed frame · Frame 7

Interval boundary estimated by the model from saved frames, not human ground truth.

Frames 17–19 · 8.00–9.20s (end exclusive)

Code indices f16–f18 (0-based, 2 FPS sampled sequence); source files 00001892.jpg → 00001907.jpg. Last observed frame time: 9.00s.

Clip end reached; the true end may be later

g0 → #4 · wheeled machine
An orange excavator / its arm / It Category check: agreeg1 → #6 · truck
a parked truck Category check: agreeg2 → #2 · ground
the dirt ground Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 3/6; subject/object links: 4; unlinked: 0.

1688_ePhZYeMtsSU

1 actionsText check: repaired3.0s · 1 model call

Frame 1 (f0) · source frame 3303 · 0.00s

Frame 2 (f1) · source frame 3309 · 0.40s

Frame 3 (f2) · source frame 3318 · 1.00s

Frame 4 (f3) · source frame 3324 · 1.40s

Frame 5 (f4) · source frame 3333 · 2.00s

Frame 6 (f5) · source frame 3339 · 2.40s

Frame 7 (f6) · source frame 3345 · 2.80s

Caption + Actions + Temporal Localization

An orange lifeboat moves rightward along the river. A smaller white boat is visible farther behind it.

movessimple · highSubject: An orange lifeboat g0 → #7

Located · Localization confidence high: The orange lifeboat advances rightward through the water throughout the sequence, trailing a visible wake.

Frame sequence:[[0, 6]]Time sequence:[[0.0, 3.0]]

Frames 1–7 · 0.00–3.00s (end exclusive)

Code indices f0–f6 (0-based, 2 FPS sampled sequence); source files 00003303.jpg → 00003345.jpg. Last observed frame time: 2.80s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #7 · ship or boat
An orange lifeboat / it Category check: unknowng1 → #4 · river
the river Category check: agreeg2 → #6 · ship or boat
A smaller white boat Category check: agree

Local validation details (entities repaired · actions ok)

entities: mentions[3]: re-anchored 'it' [64, 66] -> [99, 101] (word-bounded occurrence)

Entity coverage: 3/7; subject/object links: 1; unlinked: 0.

1822_joLWlEd2bPM

0 actionsText check: ok3.0s · 1 model call

Frame 1 (f0) · source frame 3078 · 0.00s

Frame 2 (f1) · source frame 3084 · 0.40s

Frame 3 (f2) · source frame 3093 · 1.00s

Frame 4 (f3) · source frame 3099 · 1.40s

Frame 5 (f4) · source frame 3108 · 2.00s

Frame 6 (f5) · source frame 3114 · 2.40s

Frame 7 (f6) · source frame 3120 · 2.80s

Caption + Actions + Temporal Localization

Furrowed ground and tall grass fill the foreground. A small white house is near dense trees beneath a pale sky.

No actions extracted.

g0 → #1 · ground
Furrowed ground Category check: agreeg1 → #2 · grass
tall grass Category check: unknowng2 → #3 · house
A small white house Category check: agreeg3 → #5 · tree
dense trees Category check: agreeg4 → #4 · sky
a pale sky Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 5/5; subject/object links: 0; unlinked: 0.

196_cS5frkjmPgA

4 actionsText check: ok4.2s · 1 model call

Caption + Actions + Temporal Localization

Many people skate or stand near a golden sculpture outdoors. Among them, a child glides away from a black-clad adult. Flags flutter overhead.

skatesimple · highSubject: Many people g0 → #51, g1 → #46, g2 → #45, g3 → #25, g4 → #42, g5 → #43, g6 → #29, g7 → #44, g8 → #24, g9 → #32, g10 → #26, g11 → #39, g12 → #34, g13 → #33, g14 → #48, g15 → #20, g16 → #36, g17 → #31, g18 → #37, g19 → #38, g20 → #30, g21 → #35, g22 → #23, g23 → #40, g24 → #47, g25 → #21, g26 → #27Object: a golden sculpture g27 → #7

Located · Localization confidence medium: Multiple members skate throughout the sequence, while other members remain stationary.

Frame sequence:[[0, 8]]Time sequence:[[0.0, 4.2]]

Frames 1–9 · 0.00–4.20s (end exclusive)

Code indices f0–f8 (0-based, 2 FPS sampled sequence); source files 00000631.jpg → 00000691.jpg. Last observed frame time: 4.00s.

skate first action frame — First observed frame · Frame 1

skate last action frame — Last observed frame · Frame 9

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

standposture · highSubject: Many people g0 → #51, g1 → #46, g2 → #45, g3 → #25, g4 → #42, g5 → #43, g6 → #29, g7 → #44, g8 → #24, g9 → #32, g10 → #26, g11 → #39, g12 → #34, g13 → #33, g14 → #48, g15 → #20, g16 → #36, g17 → #31, g18 → #37, g19 → #38, g20 → #30, g21 → #35, g22 → #23, g23 → #40, g24 → #47, g25 → #21, g26 → #27Object: a golden sculpture g27 → #7

Located · Localization confidence medium: Several members hold upright stationary postures throughout, rather than every member standing still.

Frame sequence:[[0, 8]]Time sequence:[[0.0, 4.2]]

Frames 1–9 · 0.00–4.20s (end exclusive)

Code indices f0–f8 (0-based, 2 FPS sampled sequence); source files 00000631.jpg → 00000691.jpg. Last observed frame time: 4.00s.

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

glidessimple · highSubject: a child g8 → #24Object: a black-clad adult g3 → #25

Located · Localization confidence high: The child moves rightward and increasingly separates from the black-clad adult.

Frame sequence:[[4, 8]]Time sequence:[[2.0, 4.2]]

Frames 5–9 · 2.00–4.20s (end exclusive)

Code indices f4–f8 (0-based, 2 FPS sampled sequence); source files 00000661.jpg → 00000691.jpg. Last observed frame time: 4.00s.

Clip end reached; the true end may be later

fluttersimple · mediumSubject: Flags g28 → #19, g29 → #49, g30 → #8, g31 → #9, g32 → #12, g33 → #11, g34 → #10, g35 → #16, g36 → #18, g37 → #15, g38 → #14, g39 → #13

Located · Localization confidence medium: Several flags visibly change their folds and outlines, although movement is subtle for some.

Frame sequence:[[0, 8]]Time sequence:[[0.0, 4.2]]

Frames 1–9 · 0.00–4.20s (end exclusive)

Code indices f0–f8 (0-based, 2 FPS sampled sequence); source files 00000631.jpg → 00000691.jpg. Last observed frame time: 4.00s.

flutter first action frame — First observed frame · Frame 1

flutter last action frame — Last observed frame · Frame 9

Clip start reached; the true start may be earlier; Clip end reached; the true end may be later

g0 → #51 · person
Many people / them Category check: agreeg1 → #46 · person
Many people / them Category check: agreeg2 → #45 · person
Many people / them Category check: agreeg3 → #25 · person
Many people / them / a black-clad adult Category check: agreeg4 → #42 · person
Many people / them Category check: agreeg5 → #43 · person
Many people / them Category check: agreeg6 → #29 · person
Many people / them Category check: agreeg7 → #44 · person
Many people / them Category check: agreeg8 → #24 · person
Many people / them / a child Category check: agreeg9 → #32 · person
Many people / them Category check: agreeg10 → #26 · person
Many people / them Category check: agreeg11 → #39 · person
Many people / them Category check: agreeg12 → #34 · person
Many people / them Category check: agreeg13 → #33 · person
Many people / them Category check: agreeg14 → #48 · person
Many people / them Category check: agreeg15 → #20 · person
Many people / them Category check: agreeg16 → #36 · person
Many people / them Category check: agreeg17 → #31 · person
Many people / them Category check: agreeg18 → #37 · person
Many people / them Category check: agreeg19 → #38 · person
Many people / them Category check: agreeg20 → #30 · person
Many people / them Category check: agreeg21 → #35 · person
Many people / them Category check: agreeg22 → #23 · person
Many people / them Category check: agreeg23 → #40 · person
Many people / them Category check: agreeg24 → #47 · person
Many people / them Category check: agreeg25 → #21 · person
Many people / them Category check: agreeg26 → #27 · person
Many people / them Category check: agreeg27 → #7 · sculpture
a golden sculpture Category check: agreeg28 → #19 · flag
Flags Category check: agreeg29 → #49 · flag
Flags Category check: agreeg30 → #8 · flag
Flags Category check: agreeg31 → #9 · flag
Flags Category check: agreeg32 → #12 · flag
Flags Category check: agreeg33 → #11 · flag
Flags Category check: agreeg34 → #10 · flag
Flags Category check: agreeg35 → #16 · flag
Flags Category check: agreeg36 → #18 · flag
Flags Category check: agreeg37 → #15 · flag
Flags Category check: agreeg38 → #14 · flag
Flags Category check: agreeg39 → #13 · flag
Flags Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 40/40; subject/object links: 7; unlinked: 0.

2112_N3YcO7YTOBE

5 actionsText check: ok3.0s · 1 model call

Caption + Actions + Temporal Localization

A man holds a handstand indoors. He lowers himself and stands upright. Two people descend on the left, and another person moves on the right.

holds a handstandlight_verb · highSubject: A man g0 → #6

Located · Localization confidence high: The man supports himself upside down through frame 3, when one leg begins dropping.

Frame sequence:[[0, 3]]Time sequence:[[0.0, 1.9]]

Frames 1–4 · 0.00–1.90s (end exclusive)

Code indices f0–f3 (0-based, 2 FPS sampled sequence); source files 00000483.jpg → 00000504.jpg. Last observed frame time: 1.40s.

holds a handstand first action frame — First observed frame · Frame 1

holds a handstand last action frame — Last observed frame · Frame 4

Clip start reached; the true start may be earlier

lowerssimple · highSubject: He g0 → #6Object: himself g0 → #6

Located · Localization confidence high: He brings his legs down from the handstand and returns to his feet.

Frame sequence:[[3, 4]]Time sequence:[[1.4, 2.4]]

Frames 4–5 · 1.40–2.40s (end exclusive)

Code indices f3–f4 (0-based, 2 FPS sampled sequence); source files 00000504.jpg → 00000513.jpg. Last observed frame time: 2.00s.

lowers first action frame — First observed frame · Frame 4

lowers last action frame — Last observed frame · Frame 5

Interval boundary estimated by the model from saved frames, not human ground truth.

standsposture · highSubject: He g0 → #6

Located · Localization confidence high: He is back on both feet and becomes upright in the final frames.

Frame sequence:[[5, 6]]Time sequence:[[2.4, 3.0]]

Frames 6–7 · 2.40–3.00s (end exclusive)

Code indices f5–f6 (0-based, 2 FPS sampled sequence); source files 00000519.jpg → 00000525.jpg. Last observed frame time: 2.80s.

Clip end reached; the true end may be later

descendsimple · highSubject: Two people g1 → #8, g2 → #7

Located · Localization confidence medium: Both people on the left become progressively lower and larger as they descend.

Frame sequence:[[2, 6]]Time sequence:[[1.0, 3.0]]

Frames 3–7 · 1.00–3.00s (end exclusive)

Code indices f2–f6 (0-based, 2 FPS sampled sequence); source files 00000498.jpg → 00000525.jpg. Last observed frame time: 2.80s.

descend first action frame — First observed frame · Frame 3

descend last action frame — Last observed frame · Frame 7

Clip end reached; the true end may be later

movessimple · mediumSubject: another person g3 → #9

Located · Localization confidence medium: The small person on the right shifts rightward across the later frames.

Frame sequence:[[3, 6]]Time sequence:[[1.4, 3.0]]

Frames 4–7 · 1.40–3.00s (end exclusive)

Code indices f3–f6 (0-based, 2 FPS sampled sequence); source files 00000504.jpg → 00000525.jpg. Last observed frame time: 2.80s.

Clip end reached; the true end may be later

g0 → #6 · person
A man / He / himself Category check: agreeg1 → #8 · person
Two people Category check: agreeg2 → #7 · person
Two people Category check: agreeg3 → #9 · person
another person Category check: agree

Local validation details (entities ok · actions ok)

No text-alignment issues.

Entity coverage: 4/9; subject/object links: 6; unlinked: 0.