Is Visual in-Context Learning for Compositional Medical Tasks within Reach?

TL;DR

This paper investigates the feasibility of visual in-context learning for complex, multi-step medical tasks, introducing a synthetic task generation method and analyzing training strategies to improve model adaptability.

Contribution

It presents a novel synthetic task generation engine and insights into training in-context learners for compositional visual tasks, especially in medical applications.

Findings

Visual in-context learning can handle multi-step tasks with proper training.

Synthetic task generation improves model adaptability to complex tasks.

Masking strategies influence training effectiveness for compositional tasks.

Abstract

In this paper, we explore the potential of visual in-context learning to enable a single model to handle multiple tasks and adapt to new tasks during test time without re-training. Unlike previous approaches, our focus is on training in-context learners to adapt to sequences of tasks, rather than individual tasks. Our goal is to solve complex tasks that involve multiple intermediate steps using a single model, allowing users to define entire vision pipelines flexibly at test time. To achieve this, we first examine the properties and limitations of visual in-context learning architectures, with a particular focus on the role of codebooks. We then introduce a novel method for training in-context learners using a synthetic compositional task generation engine. This engine bootstraps task sequences from arbitrary segmentation datasets, enabling the training of visual in-context learners for compositional tasks. Additionally, we investigate different masking-based training objectives to gather insights into how to train models better for solving complex, compositional tasks. Our exploration not only provides important insights especially for multi-modal medical task sequences but also highlights challenges that need to be addressed.