Webly Supervised Concept Expansion for General Purpose Vision Models

Benchmark Model Rank Results
object-categorization-on-gritGPV-2Categorization (ablation): 54.7Categorization (test): 55.1
object-localization-on-gritGPV-2Localization (ablation): 53.6Localization (test): 53.6
visual-question-answering-on-a-okvqaGPV-2MC Accuracy: 53.7DA VQA Score: 40.7