Datasets:

Abk802
/

ORBIT

image_id stringclasses 120 values	image_type stringclasses 3 values	theme stringclasses 26 values	reasoning_complexity stringclasses 3 values	question_id stringclasses 3 values	question stringlengths 25 197	answer int64 0 10	property_category stringclasses 4 values	objects stringlengths 1 183 ⌀
image01	AI_GENERATED	driveway	direct_recognition	Q1	How many objects made of rubber are visible?	7	physical	null
image01	AI_GENERATED	driveway	property_inference	Q2	How many objects are visible that are means of transportation?	4	functional	null
image01	AI_GENERATED	driveway	counterfactual_reasoning	Q3	If the car in the driveway were to leave, how many objects primarily made of metal would be present?	4	physical	car on the left, swing set, car in garage, bike
image02	AI_GENERATED	construction	direct_recognition	Q1	How many objects made of concrete are present?	5	physical	null
image02	AI_GENERATED	construction	property_inference	Q2	How many objects are visible that can be used for holding liquids?	2	relational	two paint cans
image02	AI_GENERATED	construction	counterfactual_reasoning	Q3	If the orange paint spilt all over one of the plexiglass sheets, how many objects would remain that are transparent?	1	physical	null
image03	AI_GENERATED	farm	direct_recognition	Q1	How many mammals are present in the image?	5	taxonomic	null
image03	AI_GENERATED	farm	property_inference	Q2	How many people are sat crouching and feeding an animal?	1	relational	null
image03	AI_GENERATED	farm	counterfactual_reasoning	Q3	If the two sheep were replaced by a cow grazing in the same area, how many objects would be present in between the two fences?	3	relational	null
image04	AI_GENERATED	library	direct_recognition	Q1	How many objects are visible that are made of paper?	7	physical	null
image04	AI_GENERATED	library	property_inference	Q2	How many objects are present that can be worn on the head?	2	functional	glasses, headphones
image04	AI_GENERATED	library	counterfactual_reasoning	Q3	If the glasses were placed inside the ceramic container, and we use this container as a dividing line between the left and right sides of the bookshelf, how many objects would be on the right side?	4	relational	headphone, three books
image05	AI_GENERATED	picnic	direct_recognition	Q1	How many objects are visible that are made of porcelain?	3	physical	null
image05	AI_GENERATED	picnic	property_inference	Q2	How many figurines are placed on the table?	3	relational	null
image05	AI_GENERATED	picnic	counterfactual_reasoning	Q3	If the drinks were split evenly between the two humans, how many drinks would each human consume?	3	relational	null
image06	AI_GENERATED	beach cleanup	direct_recognition	Q1	How many mammals are present in the image?	5	taxonomic	null
image06	AI_GENERATED	beach cleanup	property_inference	Q2	How many objects are visible that are designed to contain liquids?	8	functional	six clear bottles, two green bottles
image06	AI_GENERATED	beach cleanup	counterfactual_reasoning	Q3	If the trash bags and bottles on the sand are only thrown into the black bin, how many mammals are actively holding some other object?	3	relational	null
image07	AI_GENERATED	camping	direct_recognition	Q1	How many mammals are present in the image?	4	taxonomic	three people, one dog
image07	AI_GENERATED	camping	property_inference	Q2	How many objects are present that provide shelter?	4	functional	four tents
image07	AI_GENERATED	camping	counterfactual_reasoning	Q3	If one of the mammals douses the fire, how many objects are present that can be switched off?	4	functional	four lights
image08	AI_GENERATED	gym	direct_recognition	Q1	How many different types of gym equipment are present?	5	taxonomic	treadmills, elliptical machine, kettlebells, bench press station, resistance bands
image08	AI_GENERATED	gym	property_inference	Q2	How many exercise equipment are placed next to each other in a line?	3	relational	treadmills, elliptical machine
image08	AI_GENERATED	gym	counterfactual_reasoning	Q3	If one of the treadmills is faulty and removed from the gym, how many objects are present that convey some kind of information?	5	functional	clock, tv, two eliptical machine, one treadmill
image09	AI_GENERATED	garage	direct_recognition	Q1	How many objects made of rubber are visible in the image?	9	physical	tires
image09	AI_GENERATED	garage	property_inference	Q2	How many objects are visible that need electricity to operate?	4	functional	hydraulic lift, three tubelights
image09	AI_GENERATED	garage	counterfactual_reasoning	Q3	If one of the workers took a wrench off the table, how many objects would remain in physical contact with the table?	4	relational	service manual, toolbox, plier, wrench
image10	AI_GENERATED	kitchen	direct_recognition	Q1	How many objects are visible that are made of metal?	8	physical	three knives, pot, burner grates, stove, faucet, sink
image10	AI_GENERATED	kitchen	property_inference	Q2	How many objects present are breakable?	6	functional	five ceramic bowls, one glass
image10	AI_GENERATED	kitchen	counterfactual_reasoning	Q3	If the bowls with the tomatoes and the chickpeas were emptied into the steaming pot, how many containers would still have something remaining in them?	5	functional	two bowls green herbs, one bowl spinach, one steaming pot, one glass water
image11	AI_GENERATED	bedroom	direct_recognition	Q1	How many plants are present?	5	taxonomic	null
image11	AI_GENERATED	bedroom	property_inference	Q2	How many objects that are hanging off something are visible?	7	relational	clothes, star
image11	AI_GENERATED	bedroom	counterfactual_reasoning	Q3	If the white dress on the right is hung off the rod on the left, how many individual items are visible hanging off the rod?	6	relational	clothes
image12	AI_GENERATED	classroom	direct_recognition	Q1	How many tables are visible?	4	taxonomic	null
image12	AI_GENERATED	classroom	property_inference	Q2	How many students are looking away from the viewer?	2	relational	null
image12	AI_GENERATED	classroom	counterfactual_reasoning	Q3	If half of the total quantity of clocks on the wall are removed, how many circular objects would be visible?	1	physical	clock
image13	AI_GENERATED	kitchen	direct_recognition	Q1	How many blue-shaded objects are visible?	6	physical	jeans, bucket, mop, scrubber, cleaning fluid, towel
image13	AI_GENERATED	kitchen	property_inference	Q2	How many appliances can warm up bread?	2	functional	oven, toaster
image13	AI_GENERATED	kitchen	counterfactual_reasoning	Q3	If the bucket was full of water, how many liquid-holding objects would be visible?	2	functional	bucket, spray bottle
image14	AI_GENERATED	kitchen	direct_recognition	Q1	How many rooms are depicted in this image?	1	taxonomic	null
image14	AI_GENERATED	kitchen	property_inference	Q2	How many items can be used for cutting?	4	functional	null
image14	AI_GENERATED	kitchen	counterfactual_reasoning	Q3	If every man would be replaced by three dogs, how many mammals would there be in the image?	3	taxonomic	null
image15	AI_GENERATED	kitchen	direct_recognition	Q1	How many bowls are present?	4	taxonomic	bowls
image15	AI_GENERATED	kitchen	property_inference	Q2	How many items hold liquid?	2	functional	pot, glass
image15	AI_GENERATED	kitchen	counterfactual_reasoning	Q3	If all pots are empty, how many containers would visibly hold food?	6	functional	null
image16	AI_GENERATED	kitchen	direct_recognition	Q1	How many items with handles are present?	4	physical	3 knives, pot
image16	AI_GENERATED	kitchen	property_inference	Q2	How many utensils are on top of wooden cutting boards?	3	relational	null
image16	AI_GENERATED	kitchen	counterfactual_reasoning	Q3	If there were twice as many knives in the image, how many metal objects would be shown in total?	8	physical	1 tap, 1 pot, 6 knives
image17	AI_GENERATED	beach cleanup	direct_recognition	Q1	How many people are wearing gloves?	3	physical	people
image17	AI_GENERATED	beach cleanup	property_inference	Q2	How many bags appear to be full of collected items?	7	relational	null
image17	AI_GENERATED	beach cleanup	counterfactual_reasoning	Q3	If the trash pickers had used bins instead of plastic bags, how many bins would be visible?	8	taxonomic	containers
image18	AI_GENERATED	construction	direct_recognition	Q1	How many headwear objects are visible?	4	taxonomic	null
image18	AI_GENERATED	construction	property_inference	Q2	How many warning signs are standing upright?	2	relational	signs
image18	AI_GENERATED	construction	counterfactual_reasoning	Q3	If the wheelbarrow were yellow, how many yellow-shaded objects would be visible in the image?	7	physical	3 helmets, 2 signs, 1 wheelbarrow, safety jacket
image19	AI_GENERATED	garage	direct_recognition	Q1	How many circular rubber objects are present?	6	physical	6 tires
image19	AI_GENERATED	garage	property_inference	Q2	How many tires touch the ground?	4	relational	null
image19	AI_GENERATED	garage	counterfactual_reasoning	Q3	If the helmets were red instead of white, how many red objects would be visible?	4	physical	2 helmets, backlight, toolbox
image20	AI_GENERATED	biking	direct_recognition	Q1	How many tires are present?	7	taxonomic	null
image20	AI_GENERATED	biking	property_inference	Q2	How many objects are visible that can be used to protect the head?	2	functional	helmets
image20	AI_GENERATED	biking	counterfactual_reasoning	Q3	If the bicycle being repaired were replaced by a car, how many bicycles would be fully assembled?	0	taxonomic	null
image21	AI_GENERATED	park	direct_recognition	Q1	How many mammals are present?	3	taxonomic	fox, groundhog, squirrel
image21	AI_GENERATED	park	property_inference	Q2	How many vegetables are visible in front of the log?	3	relational	3 mushrooms
image21	AI_GENERATED	park	counterfactual_reasoning	Q3	If every book were open, how many pages would be visible?	4	physical	null
image22	AI_GENERATED	park	direct_recognition	Q1	How many mammal species are visible?	3	taxonomic	fox, rabbits, squirrels
image22	AI_GENERATED	park	property_inference	Q2	How many objects are on top of the logs?	5	relational	rabbit, fox, crystal, two squirrels
image22	AI_GENERATED	park	counterfactual_reasoning	Q3	If the mushrooms were all the same material as the crystal, how many glass objects would there be?	3	physical	crystal, two mushrooms
image23	AI_GENERATED	biking	direct_recognition	Q1	How many wheels are present?	5	taxonomic	two bikes, one wheel barrow
image23	AI_GENERATED	biking	property_inference	Q2	How many items can be used for safety in traffic?	2	functional	2 helmets
image23	AI_GENERATED	biking	counterfactual_reasoning	Q3	If all tools were removed from the photo, how many blue items would be visible?	1	physical	helmet
image24	AI_GENERATED	laboratory	direct_recognition	Q1	How many legs are depicted?	8	taxonomic	null
image24	AI_GENERATED	laboratory	property_inference	Q2	How many robots are shown on a screen?	2	relational	null
image24	AI_GENERATED	laboratory	counterfactual_reasoning	Q3	If all robots were to become humans, how many humans would there be?	7	taxonomic	null
image25	AI_GENERATED	laboratory	direct_recognition	Q1	How many objects with legs are visible?	4	physical	3 robots, 1 chair
image25	AI_GENERATED	laboratory	property_inference	Q2	How many items emit light?	1	functional	null
image25	AI_GENERATED	laboratory	counterfactual_reasoning	Q3	If a human replaced every robot, how many humans would be present?	3	taxonomic	null
image26	AI_GENERATED	classroom	direct_recognition	Q1	How many animals with tails are present?	3	physical	fox, squirrel, rabbit
image26	AI_GENERATED	classroom	property_inference	Q2	How many objects are present that provide a source of illumination?	2	functional	crystals
image26	AI_GENERATED	classroom	counterfactual_reasoning	Q3	If the crystals were replaced by squirrels, how many animals would be on top of logs?	5	relational	fox, owl, rabbit, 2 squirrels
image27	AI_GENERATED	park	direct_recognition	Q1	How many birds are visible?	2	taxonomic	owls
image27	AI_GENERATED	park	property_inference	Q2	How many birds are sitting on a tree stump?	1	relational	owl
image27	AI_GENERATED	park	counterfactual_reasoning	Q3	If there were twice as many squirrels as rabbits, how many squirrels would there be?	4	taxonomic	null
image28	AI_GENERATED	market	direct_recognition	Q1	How many hands of bananas are visible?	3	taxonomic	null
image28	AI_GENERATED	market	property_inference	Q2	How many people are interacting besides the cart?	4	relational	null
image28	AI_GENERATED	market	counterfactual_reasoning	Q3	If the carrots were replaced by tomatoes, how many cart compartments would contain red produce?	4	physical	null
image29	AI_GENERATED	tools	direct_recognition	Q1	How many liquid objects are present?	0	taxonomic	null
image29	AI_GENERATED	tools	property_inference	Q2	How many wearable items are visible?	4	functional	gloves
image29	AI_GENERATED	tools	counterfactual_reasoning	Q3	If every person wears exactly 1 glove, how many people could use the visible gloves in total?	4	relational	tools
image30	AI_GENERATED	tools	direct_recognition	Q1	How many metal tools are fully visible?	5	taxonomic	null
image30	AI_GENERATED	tools	property_inference	Q2	How many tools are designed to grip or tighten other objects?	5	functional	wrenches
image30	AI_GENERATED	tools	counterfactual_reasoning	Q3	If all orange handles were wooden, how many wooden objects would be visible?	2	physical	table, handle
image31	AI_GENERATED	library	direct_recognition	Q1	How many loose pieces of furniture are present?	5	taxonomic	dresser, desk, 2 chairs, 1 stool
image31	AI_GENERATED	library	property_inference	Q2	How many objects are on top of desks?	3	relational	book on desk, globe on dresser
image31	AI_GENERATED	library	counterfactual_reasoning	Q3	If all stools were replaced by paintings, how many paintings would there be?	3	physical	null
image32	AI_GENERATED	library	direct_recognition	Q1	How many objects have multiple legs?	4	physical	2 chairs, desk, dresser
image32	AI_GENERATED	library	property_inference	Q2	How many items hang on the walls?	2	relational	null
image32	AI_GENERATED	library	counterfactual_reasoning	Q3	If all the books were removed from the shelves, how many empty rows would be visible?	6	physical	null
image33	AI_GENERATED	kitchen	direct_recognition	Q1	How many tools are present?	4	physical	null
image33	AI_GENERATED	kitchen	property_inference	Q2	How many knives are leaning against the cutting board?	2	relational	null
image33	AI_GENERATED	kitchen	counterfactual_reasoning	Q3	If the cutting boards had handles, how many objects would handles would be present?	8	physical	2 kettles, 4 tools, 2 planks
image34	AI_GENERATED	kitchen	direct_recognition	Q1	How many shelves are present?	5	physical	null

End of preview. Expand in Data Studio

ORBIT: An Object Property Reasoning Benchmark for Visual Inference Tasks

Inspired by human object categorization, this repository introduces ORBIT, a comprehensive benchmark designed to evaluate the abilities of Vision–Language Models (VLMs) to reason about abstract object properties. ORBIT spans four object property dimensions (physical, taxonomic, functional, relational), three levels of reasoning complexity (direct recognition, property inference, counterfactual reasoning), and three visual domains (photographic/real, animated, AI-generated).

📄 Paper | 📚 arXiv | 💻 Code

Uses:

This dataset is designed to test VLMs through count-based questions, enabling evaluation of how well they reason about abstract object properties across different visual contexts.

Fields:

image_id: the image number (120 under each image_type).
image_type: the three visual domains (photographic/real, animated, AI-generated).
theme: label to indicate image theme.
reasoning_complexity: the three reasoning complexity levels (direct recognition, property inference, counterfactual).
question_id: the question number (three levels of reasoning_complexity).
question: the task for the specific level.
answer: the ground truth/correct answer to the question.
property_category: the four property dimensions (physical, taxonomic, functional, relational).
objects: list of detected objects part of ground truth (reasoning for answer).

Downloads last month: 15

Size of downloaded dataset files:

145 MB

Size of the auto-converted Parquet files:

145 MB

Number of rows:

1,080