Chop & Learn: Recognizing and Generating Object-State Compositions

TL;DR

This paper introduces a new benchmark and tasks for recognizing and generating object-state compositions, focusing on cutting objects in various styles and transferring these styles to new objects, with applications in image and video understanding.

Contribution

It presents the Chop & Learn benchmark suite and a novel Compositional Image Generation task for transferring cut styles across objects, advancing compositional understanding in vision.

Findings

Benchmark enables learning of object and cut styles from multiple viewpoints.

The generative model can transfer cut styles to new objects effectively.

Dataset supports multiple video tasks like compositional action recognition.

Abstract

Recognizing and generating object-state compositions has been a challenging task, especially when generalizing to unseen compositions. In this paper, we study the task of cutting objects in different styles and the resulting object state changes. We propose a new benchmark suite Chop & Learn, to accommodate the needs of learning objects and different cut styles using multiple viewpoints. We also propose a new task of Compositional Image Generation, which can transfer learned cut styles to different objects, by generating novel object-state images. Moreover, we also use the videos for Compositional Action Recognition, and show valuable uses of this dataset for multiple video tasks. Project website: https://chopnlearn.github.io.