Do Vision-Language Foundational models show Robust Visual Perception?
Shivam Chandhok, Pranav Tandon
TL;DR
The paper investigates whether vision-language foundational models are robust to distribution shifts by evaluating zero-shot classification under common corruptions on CIFAR10 and PASCAL VOC. It categorizes models into Contrastive Multi-Encoder, Encoder-Decoder Generative, and Hybrid, and uses a frozen visual encoder with cosine similarity to textual class descriptions to assess robustness. Results indicate hybrid models (CoCa, BLIP2) generally outperform purely contrastive approaches, and transformer-based backbones surpass ResNet-based ones in robustness, with Gaussian noise being the most challenging corruption. The findings offer practical guidance for safety-critical deployments and emphasize latent-space analyses as a tool to understand generalization, with code available at the provided GitHub repository.
Abstract
Recent advances in vision-language foundational models have enabled development of systems that can perform visual understanding and reasoning tasks. However, it is unclear if these models are robust to distribution shifts, and how their performance and generalization capabilities vary under changes in data distribution. In this project we strive to answer the question "Are vision-language foundational models robust to distribution shifts like human perception?" Specifically, we consider a diverse range of vision-language models and compare how the performance of these systems is affected by corruption based distribution shifts (such as \textit{motion blur, fog, snow, gaussian noise}) commonly found in practical real-world scenarios. We analyse the generalization capabilities qualitatively and quantitatively on zero-shot image classification task under aforementioned distribution shifts. Our code will be avaible at \url{https://github.com/shivam-chandhok/CPSC-540-Project}