Datasets:

everex
/

MVVBench

sample_name stringlengths 12 30	views listlengths 2 6	question stringlengths 22 225	choice_A stringlengths 1 168	choice_B stringlengths 1 184	choice_C stringlengths 1 127 ⌀	choice_D stringlengths 1 141 ⌀	correct_answer stringclasses 4 values	gt_answer_text stringlengths 1 184	category stringclasses 6 values	static_or_dynamic stringclasses 2 values	timestamp_start float64 0 1.64k	timestamp_end float64 0 1.65k	question_type stringclasses 3 values
sfu_basketball_05_35	[ 3, 2 ]	How many people appear in this video segment in total?	7 people	8 people	9 people	10 people	C	9 people	Counting	dynamic	1	8.8	counting
sfu_basketball_05_35	[ 1, 3 ]	When the person in the red t-shirt walks past the person in the black short-sleeved shirt and shorts, what is the person in the gray short-sleeved t-shirt doing?	Holding a basketball.	Standing up.	Holding a camera.	Sitting on a chair.	D	Sitting on a chair.	Attribute identification	dynamic	18.7	20.4	descriptive
sfu_basketball_06_16	[ 1, 4 ]	How many people appear on the screen?	4 people	5 people	6 people	7 people	A	4 people	Counting	dynamic	12.1	12.3	counting
sfu_basketball_06_16	[ 1, 3 ]	How many cameras are installed?	3 cameras	4 cameras	5 cameras	6 cameras	B	4 cameras	Counting	dynamic	3.1	3.5	counting
sfu_basketball_06_24	[ 1, 3 ]	How many people are sitting in the video?	0 people	1 person	2 people	3 people	C	2 people	Counting	dynamic	0.8	4.4	counting
sfu_basketball_06_24	[ 2, 3 ]	Where is the person playing basketball on the left in view2 located in view3?	On the left	On the right	null	null	B	On the right.	Attribute identification with view index	dynamic	5	8.8	binary
sfu_basketball_06_18	[ 1, 3 ]	How many people are there in this screen in total?	5 people	6 people	7 people	8 people	A	5 people	Counting	dynamic	0	3.3	counting
sfu_basketball_06_18	[ 1, 3 ]	How many people are standing next to the camera?	2 people	3 people	4 people	5 people	A	2 people	Counting	dynamic	2.9	5.7	counting
sfu_basketball_06_6	[ 0, 4 ]	Is no one sitting in a chair when the man in the black short-sleeved t-shirt throws the basketball?	Yes	No	null	null	B	No	Attribute identification	dynamic	5.8	9	binary
sfu_basketball_06_6	[ 1, 2 ]	How many people are in this video?	7 people	8 people	9 people	10 people	A	7 people	Counting	dynamic	6.9	8.4	counting
sfu_basketball_07_9	[ 0, 1 ]	How many people are in this video?	2 people	3 people	4 people	5 people	B	3 people	Counting	static	43.5	43.6	counting
sfu_basketball_07_9	[ 2, 4 ]	Is the person in the sky blue T-shirt closer to the person sitting on the bench or to the curtain?	the curtain	the person sitting on the bench	null	null	B	The person sitting on the bench	Relative distance	dynamic	7.1	11.3	binary
sfu_basketball_08_43	[ 2, 4 ]	How many people are wearing black t-shirts?	3 people	4 people	5 people	6 people	D	6 people	Counting	dynamic	28.8	32.5	counting
sfu_basketball_08_43	[ 0, 2 ]	Is the person wearing white pants receiving a ball from the person wearing a black sleeveless top?	Yes	No	null	null	B	No	Attribute identification	dynamic	41	41.7	binary
sfu_basketball_08_43	[ 1, 4 ]	Are the pants of the person standing with crossed legs and the shoes of the person wearing a sleeveless top the same color?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	84.6	84.6	binary
sfu_basketball_08_45	[ 4, 1 ]	How many people appear in this video segment?	6 people	7 people	8 people	9 people	D	9 people	Counting	dynamic	4.5	10.2	counting
sfu_basketball_08_45	[ 1, 4 ]	How many people are wearing white shoes?	0 people	1 person	2 people	3 people	D	3 people	Counting	dynamic	11.3	12.8	counting
sfu_basketball_08_9	[ 2, 4 ]	How many people are standing still in this video?	2 people	3 people	4 people	5 people	D	5 people	Counting	dynamic	10.1	13.2	counting
sfu_basketball_08_9	[ 1, 4 ]	How many people are wearing black long pants?	2 people	3 people	4 people	5 people	B	3 people	Counting	dynamic	23.6	26.7	counting
sfu_basketball_10_20	[ 2, 3 ]	How many people are in this video?	2 people	3 people	4 people	5 people	C	4 people	Counting	dynamic	0.2	0.6	counting
sfu_basketball_10_20	[ 2, 4 ]	How many cameras are installed?	1 camera	2 cameras	3 cameras	4 cameras	D	4 cameras	Counting	dynamic	17.6	21.7	counting
sfu_basketball_10_9	[ 2, 3 ]	When the person wearing the pink t-shirt walks over, how many people appear in the video's location?	1 person	2 people	3 people	4 people	D	4 people	Attribute identification	dynamic	5.4	8.6	counting
sfu_basketball_10_9	[ 1, 2 ]	What is the man wearing the black printed t-shirt doing?	He is browsing social media on his phone.	He is recording a video with his phone.	He is looking at the QR code on his phone.	He is talking on his phone.	C	He is looking at the QR code on his phone.	Attribute identification	dynamic	29.8	30.1	descriptive
sfu_basketball_10_31	[ 2, 4 ]	What color are the shoes commonly worn by the four people appearing in the video?	Black	White	Grey	Blue	A	Black	Attribute identification	dynamic	20.3	22.4	descriptive
sfu_basketball_10_31	[ 2, 4 ]	Is the person playing basketball standing still between the two people around the camera and the person sitting in the chair?	Yes	No	null	null	B	No	Attribute identification	dynamic	24.7	28	binary
sfu_basketball_11_26	[ 4, 2 ]	How many people appear in the video in total?	5 people	6 people	7 people	8 people	B	6 people	Counting	dynamic	6.9	12.2	counting
sfu_basketball_11_26	[ 0, 2 ]	What color are the wrist watch and shoes of the person holding the basketball?	Dark blue	Grey	White	Black	D	Black	Attribute identification	dynamic	16	16	descriptive
sfu_basketball_11_26	[ 2, 3 ]	How many of the people in the video are wearing black shoes?	5 people	6 people	7 people	8 people	A	5 people	Counting	dynamic	7.2	7.7	counting
sfu_cooking020_2	[ 1, 3 ]	Is the color of the TV table different from the color of the refrigerator?	Yes	No	null	null	B	No	Attribute identification	dynamic	1.8	3.8	binary
sfu_cooking020_2	[ 0, 2 ]	What is the man wearing black shoes doing?	He is cleaning the counter.	He is chopping vegetables.	He is peeling fruit.	He is stirring a pot on the stove.	B	He is chopping vegetables.	Attribute identification	dynamic	43.1	46.3	descriptive
sfu_cooking020_2	[ 2, 4 ]	What is the man wearing black shorts putting into the bowl?	Red onions and zucchini	Red bell peppers and green bell peppers	Strawberries and kiwis	Tomatoes and cucumbers	D	Tomatoes and cucumbers	Attribute identification	dynamic	114.5	119.9	descriptive
sfu_cooking020_2	[ 0, 1 ]	Is there a pot inside the sink with a yellow scrubber on it?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	34.2	34.9	binary
sfu_cooking023_4	[ 2, 3 ]	Is the barefoot man slicing onions?	Yes	No	null	null	B	No	Attribute identification	dynamic	0	1.9	binary
sfu_cooking023_4	[ 2, 3 ]	What is hanging below the bowl containing egg mixture?	A smartphone	A dishcloth	A kitchen rag	A hand towel	B	A dishcloth	Attribute identification	static	17.4	17.4	descriptive
sfu_cooking023_4	[ 1, 4 ]	When the person in white clothes appears on the screen, what is the man with glasses doing?	He is putting ingredients into a bowl.	He is slicing cucumbers.	He is chopping green peppers.	He is dicing tomatoes.	B	He is slicing cucumbers.	Attribute identification	dynamic	1.6	5.1	descriptive
sfu_cooking023_4	[ 1, 4 ]	Are the two people in the screen barefoot?	Yes	No	null	null	A	Yes	Attribute identification	static	46.4	46.9	binary
sfu_cooking023_8	[ 3, 2 ]	When the man holding a cell phone disappears from the screen, what is the left hand of the man filling water doing?	He is clenching his fist.	He is resting his hand on the counter.	He is holding the water tap.	He is pouring water into a glass.	A	He is clenching his fist.	Attribute identification	dynamic	0.2	5.8	descriptive
sfu_cooking023_8	[ 2, 3 ]	What is the man wearing glasses doing when the man in white clothes turns his body while looking at his phone?	He poured a drink from a bottle.	He picked up a cup and then put it down.	He picked up a kettle.	He washed a cup.	B	He picked up a cup and then put it down.	Attribute identification	dynamic	18.9	23	descriptive
sfu_cooking023_8	[ 0, 3 ]	What is the man in a short-sleeved t-shirt taking out of the drawer?	A knife.	A fork.	A spoon.	A potato peeler.	B	A fork.	Attribute identification	dynamic	70.1	71.4	descriptive
sfu_cooking027_7	[ 0, 3 ]	What is the man wearing jeans doing on this screen?	He is preparing a meal.	He is cooking eggs.	He is cleaning the stove.	He is stirring a pot of soup.	B	He is cooking eggs.	Attribute identification	dynamic	67.4	69.6	descriptive
sfu_cooking027_7	[ 1, 3 ]	Are there no eggs of the same color as the egg the man on the screen broke?	Yes	No	null	null	B	No	Attribute identification	dynamic	23.1	23.4	binary
sfu_cooking027_7	[ 1, 2 ]	How many people are wearing clothes that show their arms on this screen?	1 person	2 people	3 people	4 people	C	3 people	Counting	static	50.8	51.5	counting
sfu_cooking027_7	[ 2, 4 ]	Are the refrigerator and the kitchen stove the same color?	Yes	No	null	null	A	Yes	Attribute identification	static	66.6	66.6	binary
sfu_cooking028_8	[ 0, 1 ]	What is the man wearing glasses holding inside the plastic bag?	Bread	Egg	Cucumber	Powder	C	Cucumber	Attribute identification	dynamic	8.7	11.2	descriptive
sfu_cooking028_8	[ 0, 3 ]	What is the man in jeans doing at the sink?	He is washing his hands.	He is washing dishes.	He is rinsing vegetables.	He is cleaning the sink basin.	A	He is washing his hands.	Attribute identification	dynamic	16	19.6	descriptive
sfu_cooking028_8	[ 2, 4 ]	When the person in the white shirt leans against the sofa and stands up, is the person in the black T-shirt putting the seasoning down?	Yes	No	null	null	B	No	Attribute identification	dynamic	315	323.2	binary
sfu_cooking028_8	[ 2, 4 ]	How many people with black hair are in this video?	0 people	1 person	2 people	3 people	D	3 people	Counting	dynamic	86	88	counting
sfu_cooking029_8	[ 0, 2 ]	What is the woman wearing pink socks doing?	She is placing an egg in a tray.	She is picking up an egg.	She is reaching for a plastic bag.	She is holding an egg.	B	She is picking up an egg.	Attribute identification	dynamic	6.9	7.1	descriptive
sfu_cooking029_8	[ 4, 1 ]	Is there a cup the same color as the socks the woman in the video is wearing?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	51.6	52.7	binary
sfu_cooking029_8	[ 0, 2 ]	Is the woman wearing striped clothes's wrist bare?	Yes	No	null	null	B	No	Attribute identification	dynamic	70.8	72.1	binary
sfu_cooking031_1	[ 0, 3 ]	What is the woman in this screen taking out of the refrigerator?	Eggs	Tofu	Cheese	Milk	A	Eggs	Attribute identification	dynamic	55.7	56.9	descriptive
sfu_cooking031_1	[ 1, 2 ]	Is there a tumbler the same color as the slippers the woman is wearing?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	59.2	60.3	binary
sfu_cooking031_1	[ 2, 3 ]	Is the woman in the screen wearing plain socks and a black wristwatch?	Yes	No	null	null	B	No	Attribute identification	dynamic	254.4	256.3	binary
sfu_cooking031_1	[ 0, 4 ]	How many people are there in this screen?	1 person	2 people	3 people	4 people	D	4 people	Counting	static	51.1	51.4	counting
sfu_cooking_004_3	[ 0, 2 ]	How many people are in this video?	2 people	3 people	4 people	5 people	A	2 people	Counting	dynamic	0.9	1.7	counting
sfu_cooking_004_3	[ 3, 4 ]	Is the pattern on the leftmost towel hanging on the oven red, like a tomato?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	3.7	7	binary
sfu_cooking_004_3	[ 2, 3 ]	How many kitchen appliances are the same color as the front door?	2 items	3 items	4 items	5 items	C	4 items	Counting	dynamic	9.1	10	counting
sfu_cooking_004_3	[ 2, 1 ]	Are none of the dishes used in this video white?	Yes	No	null	null	B	No	Attribute identification	dynamic	110.2	114.9	binary
sfu_cooking_007_7	[ 1, 2 ]	How many cups are there in this video?	2 cups	3 cups	4 cups	5 cups	B	3 cups	Counting	dynamic	19.9	21.4	counting
sfu_cooking_007_7	[ 2, 3 ]	Are the refrigerator and the kitchen stove the same color?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	80.1	81.7	binary
sfu_cooking_007_7	[ 3, 4 ]	What is the person in the blue shirt doing while the person in the gray hoodie is getting water in the pot?	Moving towards the window.	Standing near the kitchen island.	Looking at the person getting water.	Taking a sip from a cup.	A	Moving towards the window.	Attribute identification	dynamic	2.5	7.6	descriptive
sfu_cooking_007_7	[ 1, 3 ]	Which is the person in the hoodie closer to, the refrigerator or the pink frying pan?	the refrigerator	the pink frying pan	null	null	B	The pink frying pan	Relative distance	dynamic	130.9	135.9	binary
sfu_cooking_008_5	[ 0, 3 ]	What is the man in shorts doing?	He is pouring water.	He is checking the contents of the pot.	He is wiping down the stovetop.	He is stirring food in a pan.	A	He is pouring water.	Attribute identification	dynamic	15	17.8	descriptive
sfu_cooking_008_5	[ 1, 4 ]	What is the man in the gray shirt doing when the man in black opens the lid of the item in his hand?	He is eating yogurt from a cup.	He is smelling the contents of the cup.	He is drinking water.	He is drinking juice.	C	He is drinking water.	Attribute identification	dynamic	2.6	8.7	descriptive
sfu_cooking_008_5	[ 3, 4 ]	Is there no electric kettle of the same color as the electric rice cooker next to the pot?	Yes	No	null	null	B	No	Attribute identification	dynamic	116.5	118.1	binary
sfu_cooking_008_5	[ 3, 4 ]	How many black plugs are plugged into the outlet on the gray tiled wall?	3	4	5	6	B	4	Counting	dynamic	140.1	143.2	counting
sfu_cooking_010_7	[ 3, 4 ]	Is there a kettle of the same color as the microwave next to the refrigerator?	Yes	No	null	null	A	Yes	Attribute identification	static	26	26.5	binary
sfu_cooking_010_7	[ 2, 4 ]	Are there bread or an egg carton missing from the dining table?	Yes	No	null	null	B	No	Attribute identification	dynamic	94.7	96.7	binary
sfu_cooking_010_7	[ 0, 4 ]	What is the man in the orange shirt doing?	He is slicing fruit on a cutting board with a knife.	He is mincing vegetables with a kitchen knife.	He is preparing a salad by arranging various ingredients.	He is chopping fresh herbs with a kitchen knife.	B	He is mincing vegetables with a kitchen knife.	Attribute identification	dynamic	95.1	96.3	descriptive
sfu_covid_004_12	[ 1, 2 ]	Into which nostril is the man in black shoes inserting the cotton swab?	Left	Right	null	null	B	Right	Attribute identification	dynamic	101.4	103.2	binary
sfu_covid_004_12	[ 0, 1 ]	Is there a ring on the left hand of the man checking the COVID test kit?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	161.9	162.4	binary
sfu_covid_004_12	[ 1, 3 ]	Considering the camera that filmed view1 as the reference, where is the camera that filmed view3 located?	It is in front.	It is located on the right.	It is behind.	It is on the left.	B	It is located on the right.	Relative pose	dynamic	0	177.6	descriptive
sfu_covid_004_6	[ 1, 2 ]	Considering the camera filming view1, where is the camera filming view2 located?	It is located further to the right.	It is located to the far left.	It is located directly in front of the man.	It is located directly in the center.	D	It is located directly in the center.	Relative pose	dynamic	24.6	29.5	descriptive
sfu_covid_004_6	[ 1, 2 ]	How many people appear on this screen?	3 people	4 people	5 people	6 people	A	3 people	Counting	dynamic	73	75	counting
sfu_covid_004_6	[ 1, 2 ]	Does the man putting a cotton swab in his right nostril have no white stripes on his shoes?	Yes	No	null	null	B	No	Attribute identification	dynamic	201.2	203.1	binary
sfu_covid_011_6	[ 1, 4 ]	How many people are standing in this video?	3 people	4 people	5 people	6 people	C	5 people	Counting	dynamic	0	1	counting
sfu_covid_011_6	[ 4, 3 ]	How many people are wearing black shoes in this scene?	0 people	1 person	2 people	3 people	C	2 people	Counting	dynamic	25	25.8	counting
sfu_covid_011_6	[ 2, 3 ]	How many people are moving in this video?	0 people	1 person	2 people	3 people	C	2 people	Counting	dynamic	33	34	counting
sfu_covid_011_6	[ 3, 4 ]	What is the person closest to the camera filming view3 holding in their hand in view4?	A tablet	A mobile phone	A small camera	A remote control	B	A mobile phone	Attribute identification with view index	dynamic	35.1	36.3	descriptive
uniandes_bouldering_003_14	[ 1, 4 ]	Relative to the camera shooting view1, in which direction is the camera shooting view4 located?	It is in front.	It is on the right side of the screen.	It is behind.	It is on the left.	B	It is on the right side of the screen.	Relative pose	dynamic	0	61.9	descriptive
uniandes_bouldering_003_14	[ 1, 4 ]	Is the person holding the hold with crossed arms extending their right leg downwards?	Yes	No	null	null	A	Yes.	Attribute identification	dynamic	57.6	58	binary
uniandes_bouldering_003_14	[ 2, 4 ]	Is the black cloth wrapped around the side and front of the rock wall?	Yes	No	null	null	A	Yes.	Attribute identification	dynamic	14.4	14.9	binary
uniandes_bouldering_003_19	[ 0, 4 ]	Is what the person in black clothes is holding in their left hand the same as what the person in the red short-sleeved t-shirt is holding?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	5.6	6.3	binary
uniandes_bouldering_003_19	[ 2, 4 ]	Is there no wall in the video that has a similar color to the fire extinguisher?	Yes	No	null	null	B	No	Attribute identification	dynamic	57.9	58.8	binary
uniandes_bouldering_003_19	[ 3, 4 ]	Between the ladder and the person standing with crossed legs, which is the climbing person closer to?	The person standing with crossed legs	The ladder	null	null	A	The person standing with crossed legs	Relative distance	dynamic	41.6	45.9	binary
uniandes_bouldering_004_10	[ 3, 4 ]	Is the woman wearing a sleeveless top with a number plate on her back wearing shorts?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	0	0.8	binary
uniandes_bouldering_004_10	[ 1, 4 ]	What color hold is the person whose right foot is on the blue hold grabbing with their left hand?	Yellow	Red	Orange	Green	C	Orange	Attribute identification	dynamic	88.9	89.3	descriptive
uniandes_bouldering_004_12	[ 2, 4 ]	Between the speaker installed on the left side of the rock wall and the orange bar, which one is closer to the person climbing?	the orange bar	the speaker installed on the left side of the rock wall	null	null	A	Orange bar	Relative distance	dynamic	20	28.8	binary
uniandes_bouldering_004_12	[ 0, 3 ]	Does the person with the ponytail not have a tattoo on their left arm?	Yes	No	null	null	B	No	Attribute identification	dynamic	29.6	30.5	binary
uniandes_bouldering_004_29	[ 1, 2 ]	Based on the camera shooting view2, in which direction is the camera shooting view1?	It is in front.	It is behind.	It is on the right.	It is on the left.	D	It is on the left.	Relative pose	dynamic	0	50	descriptive
uniandes_bouldering_004_29	[ 0, 4 ]	How many tattoos are on the left arm of the person wearing gray shorts and climbing?	0	1	2	3	C	2	Counting	dynamic	20.6	21.6	counting
uniandes_bouldering_004_33	[ 3, 4 ]	Is the person climbing closer to the person sitting on the floor than to the person hidden behind the black cloth on the left side of the rock wall?	Yes	No	null	null	A	Yes	Relative distance	dynamic	39.8	44.2	binary
uniandes_bouldering_004_33	[ 0, 1 ]	What is the woman in shorts looking at?	She is checking for her climbing partner.	She is checking the climbing wall.	She is checking the paper.	She is checking her water bottle.	C	She is checking the paper.	Attribute identification	dynamic	1.5	2	descriptive
uniandes_bouldering_004_51	[ 3, 4 ]	On this screen, do some of the people sitting downstairs not have black hair?	Yes	No	null	null	B	No	Attribute identification	dynamic	18	18.6	binary
uniandes_bouldering_004_51	[ 1, 3 ]	What kind of pants is the person with hair tied with a white band wearing?	They are wearing grey leggings.	They are wearing dark blue sweatpants.	They are wearing black leggings.	They are wearing black denim jeans.	C	They are wearing black leggings.	Attribute identification	dynamic	2.5	7.7	descriptive
uniandes_bouldering_012_13	[ 1, 4 ]	What is the man holding in his right hand, who is holding a brush in his left hand?	A harness strap.	He is not holding anything.	A chalk bag.	A small clip.	B	He is not holding anything.	Attribute identification	dynamic	5.4	6.5	descriptive
uniandes_bouldering_012_13	[ 2, 4 ]	How many people wearing red t-shirts appear in this video?	4 people	5 people	6 people	7 people	B	5 people	Counting	dynamic	29.2	30.6	counting
uniandes_bouldering_012_14	[ 1, 3 ]	Is the man in shorts wearing a number bib with 35 on his back?	Yes	No	null	null	A	Yes	Attribute identification	dynamic	28.7	30.4	binary
uniandes_bouldering_012_14	[ 2, 3 ]	How many people with long hair are there in the video?	2 people	3 people	4 people	5 people	A	2 people	Counting	dynamic	17.2	18	counting
uniandes_bouldering_012_14	[ 2, 4 ]	How many people are wearing white shoes?	0 people	1 person	2 people	3 people	C	2 people	Counting	dynamic	29.5	30.4	counting

End of preview. Expand in Data Studio

MVVBench: Benchmarking 4D Reasoning in Vision-Language Models

NeurIPS 2026 · Evaluations and Datasets Track

Hyungjin Chung^1,2*, Byeongjun Park^1*, Joonseok Lee¹, Hojun Kim¹, Jaeho Choi¹, Byung-Hoon Kim^1,3

¹EverEx ²Korea University ³Yonsei University

_{* Equal contribution}

Introduction

MVVBench is a benchmark for multi-view video reasoning built from real-world multi-camera datasets. Multi-view video understanding requires integrating spatial and temporal evidence across multiple, often non-overlapping camera streams: tracking entities as they transition between viewpoints, aligning events across time, and reasoning about latent 4D continuity rather than any single visible frame.

Every question in MVVBench is curated to be monocular-ambiguous along both the view axis and the temporal axis: each question is unanswerable from any single view in the designated input set, and the majority are further unanswerable from any single moment. A question therefore becomes uniquely solvable only by jointly reasoning across views and across time. All QA pairs are human-authored by expert annotators and pass through rigorous verification.

This dataset contains 1,323 question-answer pairs across 608 multi-view video samples. Beyond benchmarking, the paper analyzes when and why current vision-language models succeed or fail, studies inference-time elicitation strategies that yield substantial gains without retraining, and presents preliminary evidence that reinforcement learning with verifiable rewards can elicit latent multi-view competence. See the paper for details.

Repository Structure

.
├── annotations/
│   ├── README.md                 # This file
│   ├── all_questions.json        # QA annotations (original format)
│   ├── all_questions.parquet     # QA annotations (parquet format)
│   ├── video_mapping.csv         # Video file mapping (source -> target)
│   └── prepare_videos.py         # Script to prepare videos from source datasets
│
├── dataset/                      # Source datasets (user must download)
│   ├── egoexo/
│   │   └── takes/
│   │       ├── sfu_basketball_01_10/
│   │       │   └── frame_aligned_videos/
│   │       │       └── downscaled/
│   │       │           └── 448/
│   │       │               ├── aria01_1201-1.mp4
│   │       │               ├── cam01.mp4
│   │       │               └── ...
│   │       └── ...
│   │
│   ├── mmptrack/
│   │   ├── train/
│   │   │   └── videos/
│   │   │       ├── 63am/
│   │   │       │   └── cafe_shop_0/
│   │   │       │       ├── cafe_shop_0_camera1.mp4
│   │   │       │       └── ...
│   │   │       └── 64am/
│   │   └── validation/
│   │       └── videos/
│   │           └── 64pm/
│   │
│   └── panoptic/
│       ├── 160224_haggling1/
│       │   └── hdVideos_down/
│       │       ├── hd_00_00.mp4
│       │       └── ...
│       └── ...
│
└── videos/                       # Prepared videos (output of prepare_videos.py)
    ├── sfu_basketball_05_35_view0.mp4
    ├── sfu_basketball_05_35_view1.mp4
    ├── 63am_cafe_shop_0_view0.mp4
    ├── 160224_haggling1_view0.mp4
    └── ...

Annotation Files

`all_questions.json` / `all_questions.parquet`

Column	Description
`sample_name`	Video sample identifier
`views`	List of view indices used for the question (e.g., `[3, 2]`)
`question`	Question text
`choice_A`, `choice_B`, `choice_C`, `choice_D`	Multiple choice options
`correct_answer`	Correct answer key (A/B/C/D)
`gt_answer_text`	Ground truth answer text
`category`	Question category
`static_or_dynamic`	Static or dynamic scene
`timestamp_start`, `timestamp_end`	Relevant time range (seconds)
`question_type`	Type: counting, descriptive, binary

`video_mapping.csv`

Maps 3,082 video files from source datasets to standardized names:

Column	Description
`target`	Target filename (e.g., `sfu_basketball_05_35_view0.mp4`)
`sample_name`	Sample name
`view`	View number
`dataset`	Source dataset: `egoexo`, `mmptrack`, `panoptic`
`split`	For mmptrack: `train` or `validation`
`source_rel_path`	Relative path within source dataset
`source_file`	Source filename

Preparing Videos

Due to licensing restrictions, videos are not included. You must download them from the original sources and use the provided script to prepare them.

Step 1: Download Source Datasets

Ego-Exo4D (2,590 files)
- Website: https://ego-exo4d-data.org/
- Download the takes with frame_aligned_videos/downscaled/448/ videos
MMPTrack (272 files)
- Website: https://iccv2021-mmp.github.io/subpage/dataset.html
- Download both train and validation splits
CMU Panoptic (220 files)
- Website: http://domedb.perception.cs.cmu.edu/
- Download sequences with hdVideos_down/ (downscaled HD videos)

Step 2: Organize Directory Structure

Place downloaded datasets under dataset/ directory following the structure shown above.

Step 3: Run the Preparation Script

# Preview first
python annotations/prepare_videos.py \
    --dataset-root ./dataset \
    --output-dir ./videos \
    --dry-run

# Copy files
python annotations/prepare_videos.py \
    --dataset-root ./dataset \
    --output-dir ./videos

# Or create symlinks (saves disk space)
python annotations/prepare_videos.py \
    --dataset-root ./dataset \
    --output-dir ./videos \
    --symlink

Statistics

Dataset	Samples	Video Files	Categories
Ego-Exo4D	496	2,590	Basketball, cooking, bouldering, dance, music, etc.
MMPTrack	68	272	Cafe, retail, office, lobby, industry safety
CMU Panoptic	44	220	Haggling, mafia, ultimatum, social games
Total	608	3,082

Citation

If you find MVVBench useful, please cite our paper:

@inproceedings{chung2026mvvbench,
  title={MVVBench: Benchmarking 4D Reasoning in Vision-Language Models},
  author={Chung, Hyungjin and Park, Byeongjun and Lee, Joonseok and Kim, Hojun and Choi, Jaeho and Kim, Byung-Hoon},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2026},
  url={https://arxiv.org/abs/2609.30952}
}

Please also cite the original source datasets:

@article{grauman2024egoexo4d,
  title={Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectives},
  author={Grauman, Kristen and others},
  journal={CVPR},
  year={2024}
}

@inproceedings{han2021mmptrack,
  title={MMPTrack: Large-scale Densely Annotated Multi-camera Multiple People Tracking Benchmark},
  author={Han, Xiaotian and others},
  booktitle={ICCV Workshop},
  year={2021}
}

@inproceedings{joo2015panoptic,
  title={Panoptic Studio: A Massively Multiview System for Social Motion Capture},
  author={Joo, Hanbyul and others},
  booktitle={ICCV},
  year={2015}
}