Robustness Analysis on Foundational Segmentation Models

Due to the increase in computational resources and accessibility of data, an increase in large, deep learning models trained on copious amounts of data using self-supervised or semi-supervised learning have emerged. These "foundation" models are often adapted to a variety of downstream tasks like classification, object detection, and segmentation with little-to-no training on the target dataset. In this work, we perform a robustness analysis of Visual Foundation Models (VFMs) for segmentation tasks and compare them to supervised models of smaller scale. We focus on robustness against real-world distribution shift perturbations.We benchmark four state-of-the-art segmentation architectures using 2 different datasets, COCO and ADE20K, with 17 different perturbations with 5 severity levels each. We find interesting insights that include (1) VFMs are not robust to compression-based corruptions, (2) while the selected VFMs do not significantly outperform or exhibit more robustness compared to non-VFM models, they remain competitively robust in zero-shot evaluations, particularly when non-VFM are under supervision and (3) selected VFMs demonstrate greater resilience to specific categories of objects, likely due to their open-vocabulary training paradigm, a feature that non-VFM models typically lack. We posit that the suggested robustness evaluation introduces new requirements for foundational models, thus sparking further research to enhance their performance.

Probing Conceptual Understanding of Large Visual-Language Models

We present a novel framework for probing and improving relational, compositional and contextual understanding of large visual-language models (V+L). While large V+L models have achieved success in various downstream tasks, it is not clear if they have a conceptual grasp of the content. We propose a novel benchmarking dataset for probing three aspects of content understanding. Our probes are grounded in cognitive science and help determine if a V+L model can, for example, determine if snow garnished with a man is implausible, or if it can identify beach furniture by knowing it is located on a beach. We have experimented with 5 well known models, such as CLIP and ViLT, and found that they mostly fail to demonstrate a conceptual understanding. That said, we find interesting insights such as cross-attention helps learning conceptual understanding. We use these insights to propose a new finetuning technique that rewards the three conceptual understanding measures we proposed. We hope that the presented benchmarks will help the community assess and improve the conceptual understanding capabilities of large V+L models.