Datasets:

Modalities:
Image
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
pandas
License:
question
stringlengths
33
117
demonstration_type
stringclasses
3 values
variation
dict
motion_type
stringclasses
1 value
answer
int64
0
5
note
stringclasses
42 values
key
stringlengths
7
7
options
sequencelengths
5
6
video_source_url
stringclasses
3 values
How many trapezoid(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
2
Eason drawing trapezoid
0209-03
[ "2", "1", "3", "4", "5", "0" ]
How many triangle(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
2
Eason drawing triangle
0210-03
[ "3", "5", "4", "1", "2", "0" ]
How many square(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
3
Eason drawing square
0211-03
[ "1", "0", "3", "2", "5", "4" ]
How many infinity-shaped curve(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
3
Eason drawing infinity-shaped curve
0212-03
[ "1", "0", "6", "7", "2", "4" ]
How many diamond shape(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
0
Eason drawing diamond
0213-03
[ "3", "0", "1", "5", "4", "2" ]
How many figure-eight shape(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
4
Eason drawing figure-eight
0214-03
[ "2", "3", "0", "1", "6", "5" ]
How many circle(s)) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
1
Eason drawing circle
0215-05
[ "5", "6", "3", "2", "0", "1" ]
How many spin(s) does the person make throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
0
Eason spinning
0216-06
[ "4", "5", "3", "1", "2", "0" ]
How many spin(s) does the person make throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
5
Eason spinning
0216-07
[ "5", "1", "3", "0", "2", "4" ]
How many circle(s)) does the person's hand trace in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
5
Eason tracing circle(s)
0218-06
[ "5", "0", "1", "7", "3", "8" ]
How many circle(s)) does the person's hand trace in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
3
Eason tracing circle(s)
0218-07
[ "4", "2", "0", "3", "5", "1" ]
How many times does the person swing their hand upward throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
5
Eason swinging hand upward
0220-01
[ "1", "3", "2", "4", "0", "5" ]
How many times does the person swing their hand to the right throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
4
Eason swinging hand to the right
0221-03
[ "2", "0", "1", "5", "6", "4" ]
How many times does the person swing their hand to the right throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
4
Eason swinging hand to the right
0222-01
[ "1", "0", "5", "8", "9", "3" ]
How many times does the person swing their hand to the right throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
0
Eason swinging hand to the right
0223-03
[ "6", "8", "5", "7", "3", "0" ]
How many times does the person swing their hand to the right throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
0
Eason swinging hand to the right with markerpen
0224-03
[ "7", "6", "3", "5", "0", "8" ]
How many times does the person swing their hand downward throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
4
Eason swinging hand downward with markerpen
0225-03
[ "0", "5", "3", "10", "9", "7" ]
How many times does the person swing their hand to the left throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
0
Eason swinging hand to the left with markerpen
0226-02
[ "7", "5", "10", "0", "9", "3" ]
How many times does the person swing their hand to the left throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
4
Eason swinging hand to the left
0227-03
[ "5", "1", "4", "3", "2", "0" ]
How many times does the person swing their hand downward throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
0
Eason swinging hand downward
0228-03
[ "4", "5", "2", "3", "1", "0" ]
How many times does the person swing their hand upward throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
0
Eason swinging hand upward
0229-03
[ "4", "3", "5", "2", "1", "0" ]
How many circle(s)) does the person's hand trace in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 1 }
count
4
Eason tracing circle(s)
0230-07
[ "1", "3", "0", "4", "5", "2" ]
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0000-00
[ "2", "6", "3", "4", "7", "5" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0001-00
[ "3", "5", "6", "2", "7", "4" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0002-00
[ "2", "1", "4", "6", "7", "5" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0003-00
[ "5", "1", "4", "3", "2", "6" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0004-00
[ "4", "1", "0", "2", "3", "5" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0005-00
[ "5", "3", "4", "7", "2", "6" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0006-00
[ "4", "3", "1", "6", "2", "5" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0007-00
[ "5", "0", "2", "4", "3", "1" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0008-00
[ "8", "1", "10", "4", "2", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
Simulated objects colliding
0009-00
[ "7", "3", "11", "5", "9", "1" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0010-00
[ "3", "4", "6", "1", "5", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0011-00
[ "3", "4", "6", "5", "8", "7" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0012-00
[ "4", "3", "2", "1", "6", "5" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0013-00
[ "6", "8", "2", "5", "4", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0014-00
[ "4", "6", "3", "1", "2", "5" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
Simulated objects colliding
0015-00
[ "4", "6", "1", "2", "5", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0016-00
[ "3", "5", "7", "6", "4", "8" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0017-00
[ "5", "9", "1", "4", "0", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
Simulated objects colliding
0018-00
[ "2", "4", "6", "3", "5", "0" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
Simulated objects colliding
0019-00
[ "0", "2", "5", "1", "4", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0020-00
[ "2", "3", "5", "0", "1", "4" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0021-00
[ "3", "5", "4", "1", "6", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0022-00
[ "1", "0", "2", "4", "3", "5" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0023-00
[ "0", "3", "1", "4", "5", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0024-00
[ "3", "1", "5", "0", "4", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0025-00
[ "3", "1", "4", "5", "2", "0" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0026-00
[ "4", "6", "3", "7", "5", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
Simulated objects colliding
0027-00
[ "0", "3", "8", "2", "1", "4" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0028-00
[ "3", "2", "6", "0", "4", "1" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0029-00
[ "3", "6", "4", "7", "2", "5" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0030-00
[ "4", "1", "0", "2", "3", "8" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0031-00
[ "0", "5", "3", "4", "1", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
Simulated objects colliding
0032-00
[ "3", "2", "0", "4", "5", "1" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0033-00
[ "3", "1", "5", "4", "0", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
Simulated objects colliding
0034-00
[ "5", "2", "0", "3", "1", "4" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0035-00
[ "0", "3", "2", "5", "4", "1" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0036-00
[ "1", "5", "4", "2", "3", "6" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0037-00
[ "0", "1", "5", "2", "4", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0038-00
[ "0", "2", "3", "5", "4", "1" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0039-00
[ "0", "4", "5", "3", "2", "1" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0040-00
[ "7", "8", "6", "5", "4", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
Simulated objects colliding
0041-00
[ "3", "5", "0", "2", "1", "4" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
Simulated objects colliding
0042-00
[ "1", "3", "5", "0", "4", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
Simulated objects colliding
0043-00
[ "4", "5", "3", "1", "0", "2" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
Simulated objects colliding
0044-00
[ "8", "2", "0", "3", "1", "4" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0045-00
[ "10", "12", "1", "8", "2", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0046-00
[ "5", "1", "0", "2", "4", "3" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
Simulated objects colliding
0047-00
[ "4", "3", "5", "1", "2", "0" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
Simulated objects colliding
0048-00
[ "3", "4", "5", "1", "2", "0" ]
CLEVRER
How many distinct collision(s) occur throughout the entire video?
simulated
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
Simulated objects colliding
0049-00
[ "7", "4", "5", "1", "3", "2" ]
CLEVRER
How many complete circle(s)) does the car make throughout the entire video?
object
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
car;rotation
0200-00
[ "1", "2", "5", "0", "4", "3" ]
How many complete circle(s) does the car make throughout the entire video?
object
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
car;static
0201-00
[ "0", "3", "5", "4", "1", "2" ]
How many complete circle(s) does the car make throughout the entire video?
object
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
car;rotation
0202-00
[ "4", "5", "3", "1", "2", "0" ]
How many complete circle(s) does the car make throughout the entire video?
object
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
car;rotation
0203-00
[ "5", "4", "2", "3", "1", "0" ]
How many complete circle(s) does the car make throughout the entire video?
object
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
car;rotation
0204-00
[ "1", "4", "2", "3", "5", "0" ]
How many spin(s) does the car make throughout the entire video?
object
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
car;rotation
0205-00
[ "4", "0", "3", "2", "6", "8" ]
How many spin(s) does the car make throughout the entire video?
object
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
car;rotation
0206-00
[ "7", "2", "9", "0", "5", "3" ]
From the person's perspective, how many spin(s) do they make throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 1, "zoom": 0 }
count
1
human;rotation
0207-00
[ "5", "2", "0", "3", "4", "1" ]
From the person's perspective, how many spin(s) do they make throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 1, "zoom": 0 }
count
4
human;rotation
0208-00
[ "1", "3", "5", "0", "2", "4" ]
How many trapezoid(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
human;drawing
0209-00
[ "4", "1", "2", "0", "5", "3" ]
How many triangle(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
human;drawing
0210-00
[ "4", "3", "1", "0", "5", "2" ]
How many square(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
human;drawing
0211-00
[ "3", "4", "5", "0", "1", "2" ]
How many infinity-shaped curve(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
human;drawing
0212-00
[ "6", "7", "4", "2", "0", "1" ]
How many diamond shape(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
human;drawing
0213-00
[ "0", "4", "2", "3", "1", "5" ]
How many figure-eight shape(s) does the person draw in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
human;drawing
0214-00
[ "2", "3", "0", "5", "1", "6" ]
How many circle(s) does the person's hand trace in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
2
human;rotation
0215-00
[ "3", "5", "6", "1", "2", "0" ]
How many spin(s) does the person make throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
human;rotation
0216-00
[ "5", "1", "3", "4", "2", "0" ]
How many spin(s) does the person make throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
human;rotation
0216-05
[ "1", "5", "0", "2", "3", "4" ]
How many circle(s) does the person's hand trace in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
human;rotation
0218-00
[ "5", "1", "7", "0", "8", "3" ]
How many circle(s) does the person's hand trace in the air throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
human;rotation
0218-01
[ "4", "3", "1", "5", "0", "2" ]
How many times does the person swing their hand upward throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
5
human;linear;markerpen
0220-00
[ "2", "0", "1", "4", "3", "5" ]
How many times does the person swing their hand to the right throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
4
human;linear
0221-00
[ "0", "1", "5", "2", "6", "4" ]
How many times does the person swing their hand to the right throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
human;linear
0222-00
[ "3", "8", "1", "9", "5", "0" ]
How many times does the person swing their hand to the right throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
1
human;linear
0223-00
[ "3", "6", "0", "8", "5", "7" ]
How many times does the person swing their hand to the right throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
human;linear;markerpen
0224-00
[ "0", "6", "3", "7", "8", "5" ]
How many times does the person swing their hand downward throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
0
human;linear;markerpen
0225-00
[ "9", "0", "10", "5", "7", "3" ]
How many times does the person swing their hand to the left throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
human;linear;markerpen
0226-00
[ "9", "0", "3", "7", "5", "10" ]
How many times does the person swing their hand to the left throughout the entire video?
human
{ "composite": 0, "counterfactual": 0, "first_person": 0, "zoom": 0 }
count
3
human;linear
0227-00
[ "1", "5", "4", "2", "3", "0" ]

πŸ… TOMATO

πŸ“„ Paper | πŸ’» Code | 🎬 Videos

This repository contains the QAs of the following paper:

πŸ… TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models
Ziyao Shangguan*1Chuhan Li*1Yuxuan Ding1Yanan Zheng1Yilun Zhao1Tesca Fitzgerald1Arman Cohan12
*Equal contribution.
1Yale University  2Allen Institute of AI

TOMATO - A Visual Temporal Reasoning Benchmark

figure1

Introduction

Our study of existing benchmarks shows that visual temporal reasoning capabilities of Multimodal Foundation Models (MFMs) are likely overestimated as many questions can be solved by using a single, few, or out-of-order frames. To systematically examine current visual temporal reasoning tasks, we propose three principles with corresponding metrics: (1) Multi-Frame Gain, (2) Frame Order Sensitivity, and (3) Frame Information Disparity.

Following these principles, we introduce TOMATO, a novel benchmark crafted to rigorously assess MFMs' temporal reasoning capabilities in video understanding. TOMATO comprises 1,484 carefully curated, human-annotated questions spanning 6 tasks (i.e. action count, direction, rotation, shape&trend, velocity&frequency, and visual cues), applied to 1,417 videos, including 805 self-recorded and -generated videos, that encompass 3 video scenarios (i.e. human-centric, real-world, and simulated). In the 805 self-created videos, we apply editing to incorporate counterfactual scenes, composite motions, and zoomed-in views, aiming to investigate the impact of these characteristics on the performance of MFMs.

Task Examples

rotation

What direction(s) does the Ping Pong ball rotate in?
A. Clockwise throughout.
B. No rotation.
C. Clockwise then counter-clockwise.
D. Counter-clockwise throughout.
E. Counter-clockwise then clockwise.

Answer: D. Counter-clockwise throughout.

acceleration

What is the pattern of the object’s speed in the video?
A. Not moving at all.
B. Constant speed.
C. Decelerating.
D. Accelerating.

Answer: C. Decelerating.

human_gesture

What instruction did the person give to the camera in the video?
A. Moving Down.
B. Moving Left.
C. Moving Further.
D. Moving Closer.
E. Moving Right.
F. Moving Up.

Answer: E. Moving Right.

synthetic_human

How many triangle(s) does the person draw in the air throughout the entire video?
A. 0
B. 1
C. 2
D. 3
E. 4
F. 5

Answer: E. 4

Analysis Highlight

earth_moon_frames

Our in-depth error case analysis reveals that models lack the basic ability to interpret frames as a continuous sequence. In the example, while GPT-4o correctly generates captions for each consecutive change in the moon's movement, showcasing its ability to reason at individual time steps, it still fails to infer based on the captions that the overall sequence represents a clockwise rotation and falsely concludes that it is a counter-clockwise rotation.

For more detailed error case analysis, please refer to Section 6.3 in our paper.

Dataset and Evaluation

1. Setup

git clone https://github.com/yale-nlp/TOMATO
cd TOMATO

Download the videos and unzip into the /TOMATO directory

After downloading the videos, your file structure should look like this.
.
β”œβ”€β”€ data/
β”œβ”€β”€ src/
β”œβ”€β”€ videos/
β”‚   β”œβ”€β”€ human/
β”‚   β”œβ”€β”€ object/
β”‚   β”œβ”€β”€ simulated/

1.1 Proprietary Models

To install the required packages for evaluating proprietary models, run:

pip install openai # GPT 
pip install google-generativeai # Gemini 
pip install anthropic # Claude
pip install reka-api==2.0.0 # Reka

Create a .env file in the root directory with the following format:

OPENAI_API_KEY="your_openai_api_key"
GEMINI_API_KEY="your_gemini_api_key"
ANTHROPIC_API_KEY="your_anthropic_api_key"
REKA_API_KEY="your_reka_api_key"

1.2 Open-sourced Models

Create a directory named pretrained in the root of TOMATO to store open-sourced models. For example, to download Qwen-2-VL-7B model, run the following command:

mkdir pretrained && cd pretrained
huggingface-cli download 
  --resume-download 
  --local-dir-use-symlinks False Qwen/Qwen2-VL-7B-Instruct 
  --local-dir Qwen2-VL-7B-Instruct
After downloading open-sourced models, your file structure should look like this.
.
β”œβ”€β”€ data/
β”œβ”€β”€ src/
β”œβ”€β”€ videos/
β”œβ”€β”€ pretrained/
β”‚   β”œβ”€β”€ Qwen2-VL-7B-Instruct/
β”‚   β”œβ”€β”€ ...

Note: To use Video-CCAM, LLaVA-NeXT, Video-LLaVA, VideoLLaMA2, and VILA, follow additional instructions below.
Clone their repositories into the ./src/generate_lib/ directory. Run the following commands:

cd ./src/generate_lib

git clone git@github.com:QQ-MM/Video-CCAM.git             # Video-CCAM
git clone git@github.com:LLaVA-VL/LLaVA-NeXT.git          # LLaVA-NeXT
git clone git@github.com:DAMO-NLP-SG/VideoLLaMA2.git      # VideoLLaMA2
git clone git@github.com:PKU-YuanGroup/Video-LLaVA.git    # Video-LLaVA
git clone git@github.com:NVlabs/VILA.git                  # VILA

After cloning, rename the directories by replacing hyphens (-) with underscores (_):

mv Video-CCAM Video_CCAM
mv LLaVA-NeXT LLaVA_NeXT
mv Video-LLaVA Video_LLaVA

2. Evaluation

To run evaluation with a model:

python src/evaluate.py 
  --model $model_name
  --reasoning_type ALL 
  --demonstration_type ALL 
  --total_frames $total_frames

All supported models are listed here. To evaluate additional models, please refer to the next section.

This is a list of models that take in videos directly and any specified total_frames will be ignore.

You can specify a subset of reasoning_type and demonstration_type using a comma-seperated list. These are the lists of valid choices.

3. Result Parsing

When our standard parser using regular expression fails, we employ GPT-4o-mini to extract answers from model response. To use the parser:

python src/parse_result.py

Note: This parser is designed to be incremental. It only parses additional raw model responses while leaving the already parsed results unchanged.

4. Display Categorized Scores

Scores are grouped by model, reasoning_type+model, and demonstration_type+model. To display scores:

python src/get_categorized_score.py

Evaluate Additional Models

Our evaluation scripts are designed for the ease of adding additional models, simply:

1. Add Model to Config File

Add model_family and model_name to src/config.json like below:

{
    "models": {
        "{model_family}": [
            "{model_name}",
            "..."
        ]

2. Add Model Evaluation Code

Create the corresponding {model_family}.py file under src/generate_lib with the starter code below:

from generate_lib.constant import GENERATION_TEMPERATURE, GENERATION_TOP_P, SYSTEM_PROMPT, MAX_TOKENS, GENERATION_SEED
from generate_lib.construct_prompt import construct_prompt
from generate_lib.utils import read_video

def generate_response(model_name: str, queries: list, total_frames: int, output_dir: str):
    # initialize your model 
    model = ...

    for query in queries:
      id_ = query['id']
      question = query['question']
      gt = optionized_list[query['answer']]

      # construct prompt
      base64Frames, _ = read_video(video_path=video_path, total_frames=total_frames)
      prompt, all_choices, index2ans = construct_prompt(question=question,
                                                        options=options,
                                                        num_frames=total_frames)
      
      # generate response
      response = model(...)

      # save model response in default format to use our result parser
      with open(output_dir, "a") as f:
            f.write(json.dumps(
                {
                    "id": id_,
                    "question": question,
                    "response": response,
                    "all_choices": all_choices,
                    "index2ans": index2ans,
                    'gt': gt
                }
            ) + "\n")

Experiments

1. Comparison with Existing Benchmarks

1.1 Multi-Frame Gain ($\kappa$): a higher value indicates the task is less solvable by a single frame.

multi_frame_gain1 multi_frame_gain2

1.2 Frame Order Sensitivity ($\tau$): a higher value indicates the task is more reliant on the correct order of frames.

frame_order_sensitivity

1.3 Frame Information Parity ($\rho$): a lower value indicates information is more evenly distributed across the frames.

frame_information_parity

2. Leaderboard

We evaluate general-purpose MFMs on TOMATO, with all models tested in a zero-shot setting. The scores below are represented percentage accuracy (%).

main_results

Contact

If you have any questions or suggestions, please don't hesitate to let us know. You can post an issue on this repository, or contact us directly at:

Citation

If you find πŸ…TOMATO useful for your research and applications, please cite using this BibTex:

@misc{shangguan2024tomatoassessingvisualtemporal,
      title={TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models}, 
      author={Ziyao Shangguan and Chuhan Li and Yuxuan Ding and Yanan Zheng and Yilun Zhao and Tesca Fitzgerald and Arman Cohan},
      year={2024},
      eprint={2410.23266},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2410.23266}, 
}
Downloads last month
471
Edit dataset card