VCR

The VCR (Visual Commonsense Reasoning) dataset is used to evaluate a model's reasoning ability in understanding the semantics of visual content and natural language through tasks that require fine-grained visual and textual information.

Papers using VCR (12)

CoRGI: Verified Chain-of-Thought Reasoning with Post-hoc Visual Grounding2025Coherent Multimodal Reasoning with Iterative Self-Evaluation for Vision-Language Models2025UniFine: A Unified and Fine-grained Approach for Zero-shot Vision-Language Understanding2023 · 2 citesVisualBERT: A Simple and Performant Baseline for Vision and Language2019 · 1,236 citesVL-BERT: Pre-training of Generic Visual-Linguistic Representations2019 · 785 citesKVL-BERT: Knowledge Enhanced Visual-and-Linguistic BERT for Visual Commonsense Reasoning2020 · 6 citesIdealGPT: Iteratively Decomposing Vision and Language Reasoning via Large Language Models2023 · 4 citesOn Advances in Text Generation from Images Beyond Captioning: A Case Study in Self-Rationalization2022Understanding ME? Multimodal Evaluation for Fine-grained Visual Commonsense2022Improving Vision-and-Language Reasoning via Spatial Relations Modeling2023From Wrong To Right: A Recursive Approach Towards Vision-Language Explanation2023Do Vision-Language Transformers Exhibit Visual Commonsense? An Empirical Study of VCR2024