ReasonCLIP-58M: Visually Grounded Commonsense Reasoning Supervision for CLIP
About
CLIP and its variants are widely adopted visual backbones in multimodal systems, but their pretraining remains dominated by descriptive image-text alignment. As downstream applications increasingly demand visually grounded commonsense inference and compositional reasoning, it remains unclear whether CLIP-style encoders can support such reasoning without architectural changes. To address this, we present ReasonCLIP-58M, a continual pretraining framework that integrates large-scale reasoning supervision into CLIP-style models through our two-stage strategy, which progressively integrates reasoning signals while preserving descriptive alignment, followed by category-structured reasoning supervision. To support this framework, we construct two complementary datasets and a benchmark: ReasonLite-42M, with open-form, visually verifiable reasoning captions; ReasonPro-16M, with category-specific reasoning supervision; and RCLIP-Bench for diagnostic evaluation of visually grounded reasoning. We train a family of ReasonCLIP that improves visually grounded commonsense and compositional reasoning while also enhancing zero-shot retrieval performance. As a drop-in visual encoder for multimodal large language models such as LLaVA-NeXT, ReasonCLIP delivers consistent gains without additional inference cost, demonstrating that structured reasoning supervision enhances the expressive capacity of CLIP-style visual representations. All datasets, models, and training code are available at https://github.com/RISys-Lab/ReasonCLIP.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Text-to-Image Retrieval | Flickr30k (test) | Recall@140.9 | 528 | |
| Compositional Reasoning | SugarCrepe | Overall Accuracy88.5 | 95 | |
| Compositional Reasoning | VALSE | Average Score79.2 | 65 | |
| Image-to-Text Retrieval | COCO 5K (test) | R@156.2 | 63 | |
| Commonsense Reasoning | WinoGAViL | Accuracy (5-6)65.2 | 22 | |
| Compositional Reasoning | WhatsUp | Accuracy51.3 | 21 | |
| Compositional Reasoning | CREPE | Accuracy44.9 | 21 | |
| Image-to-Text Compositional Reasoning | SugarCrepe++ ITT | Accuracy73.1 | 21 | |
| Text-to-Image Compositional Reasoning | SugarCrepe++ TOT | Accuracy72.9 | 21 | |
| Image-to-Text Retrieval | Urban1k (test) | R@178.6 | 9 |