Probing Task Benchmarks

AbstractJoint vision-language models have shown great performance over a diverse set of tasks. However, little is known about their limitations, as the high dimensional space learned by these models makes …...