Visual Commonsense Reasoning Q→A on VCR (val)
79.4AccuracyMERLOT
Evaluation Results
| Method | Links | |
|---|---|---|
| MERLOTLearning Paradigm=Supervised, Validation set=full validation set2023.10 | 79.4 | |
| R2CLearning Paradigm=Supervised2023.10 | 67.3 | |
| ViCorLearning Paradigm=In-Context Learning, Backbone LLM=GPT-42023.10 | 59.8 | |
| GPT-4Learning Paradigm=In-Context Learning, Reasoning Strategy=Chain-of-Thought (CoT)2023.10 | 57.8 | |
| ViCorLearning Paradigm=In-Context Learning, Backbone LLM=GPT-3.52023.10 | 55.4 | |
| BLIP2-PretrainLearning Paradigm=In-Context Learning2023.10 | 51.2 | |
| IdealGPTLearning Paradigm=In-Context Learning, Backbone LLM=GPT-3.52023.10 | 47.9 | |
| GPT-3.5Learning Paradigm=In-Context Learning, Reasoning Strategy=Chain-of-Thought (CoT)2023.10 | 43.8 |