CRAFT: A Benchmark for Causal Reasoning About Forces and inTeractions
Abstract
Recent advances in Artificial Intelligence and deep learning have revived the interest in studying the gap between the reasoning capabilities of humans and machines. In this ongoing work, we introduce CRAFT, a new visual question answering dataset that requires causal reasoning about physical forces and object interactions. It contains 38K video and question pairs that are generated from 3K videos from 10 different virtual environments, containing different number of objects in motion that interact with each other. Two question categories from CRAFT include previously studied descriptive and counterfactual questions. Besides, inspired by the theory of force dynamics from the field of human cognitive psychology, we introduce new question categories that involve understanding the intentions of objects through the notions of cause, enable, and prevent. Our preliminary results demonstrate that even though these tasks are very intuitive for humans, the implemented baselines could not cope with the underlying challenges.
Example simulations and questions
Overview
Q: “How many objects fall to the ground?” A: “2”Q: “After entering the basket, does the small yellow square collide with other objects?” A: “True”
CounterfactualQ: “How many objects fall to the ground, if the small yellow box is removed?” A: “1”Q: “Does the small gray box enter the basket, if any other single one of the objects is removed?” A: “True”
CauseQ: “Does the small brown sphere cause the tiny yellow box to enter the basket?” A: “True”Q: “Does the small gray block cause the big cyan triangle to collide with the basket?” A: “True”
EnableQ: “Does the small brown sphere enable the small yellow box to enter the basket?” A: “False”Q: “How many objects does the small gray block enable to enter the basket?” A: “0”
PreventQ: “Does the small yellow square prevent the tiny brown circle from entering the basket?” A: “True”Q: “How many objects does the large cyan triangle prevent from entering the basket?” A: “1”
Descriptive
Q: What is the color of object that the tiny cyan circle first collides with?A: blue
Q: After falling into the bucket, does the large blue block collide with other objects?A: True
Q: How many objects collide with the container?A: 3
Q: After entering the bucket, does the small brown cube collide with other objects?A: True
Q: Before getting into the container, does the large green circle collide with other objects?A: True
Q: What shape is the object that the small green triangle last collides with?A: cube
Q: After falling into the container, does the large gray circle collide with other objects?A: False
Q: Before hitting the bottom, does the tiny cyan triangle collide with other objects?A: False
Counterfactual
Q: Does the tiny green circle fall to the ground, if any other single one of the objects is removed?A: True
Q: If any other single one of the objects is removed, does the tiny green circle collide with the basket?A: False
Q: If any other single one of the objects is removed, does the small gray circle fall to the bottom?A: True
Q: If any other single one of the objects is removed, does the big gray circle enter the basket?A: True
Q: Does the big gray circle collide with the container, if any other single one of the objects is removed?A: True
Enable
Prevent
Q: Is the tiny brown circle is blocked by the small green circle from colliding with the container?A: True