Research map / Vision-language and multimodal / Multimodal LLMs
DWIM: Towards Tool-aware Visual Reasoning via Discrepancy-aware Workflow Generation & Instruct-Masking Tuning
Fucai Ke, Vijay Kumar B G, Xingjian Leng, Zhixi Cai, Zaid Khan, Weiqing Wang, Pari Delir Haghighi, Hamid Rezatofighi, Manmohan Chandraker
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add discrepancy-aware training workflow generation (ablation) | LLaMa3.1-8B generator · GQA | data utilization: 48.2 -> 68.3 | Improved |
| add discrepancy-aware training workflow generation (ablation) | DWIM · GQA | accuracy: 57.9 -> 69.3 | Improved |
| add discrepancy-aware training workflow generation (ablation) | LLaMa3.1-8B generator · SugarCREPE | data utilization: 56.5 -> 76.4 | Improved |
| add discrepancy-aware training workflow generation (ablation) | DWIM · SugarCREPE | accuracy: 60.0 -> 74.6 | Improved |
| use GPT4O instead of LLaMa3.1 as standard workflow generator (ablation) | standard generation · GQA | accuracy: 57.9 -> 58.0 | No clear effect |
| replace SFT with Instruct-Masking on standard workflows (ablation) | LLaMa3.1-8B · GQA | accuracy: 53.6 -> 57.9 | Improved |
| replace SFT with Instruct-Masking on standard workflows (ablation) | LLaMa3.1-8B · SugarCREPE | accuracy: 57.1 -> 60.0 | Improved |
| replace SFT with Instruct-Masking on discrepancy-aware workflows (ablation) | LLaMa3.1-8B · GQA | accuracy: 54.8 -> 69.3 | Improved |
| replace SFT with Instruct-Masking on discrepancy-aware workflows (ablation) | LLaMa3.1-8B · SugarCREPE | accuracy: 58.3 -> 74.6 | Improved |
| mask only effective actions instead of random masking (ablation) | Instruct-Masking · GQA | accuracy: 65.1 -> 69.3 | Improved |
| mask only effective actions instead of also masking rethink actions (ablation) | Instruct-Masking · SugarCREPE | accuracy: 70.7 -> 74.6 | Improved |
| add DWIM training to frozen 10-shot Mistral-v0.3-7B | Mistral-v0.3-7B · VCR | accuracy: 52.0 -> 67.5 | Improved |
Closest papers by meaning
- Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.90)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.90)
- CodeDance: A Dynamic Tool-integrated MLLM for Executable Visual Reasoning (CVPR 2026 · similarity 0.90)
- AdaReasoner: Dynamic Tool Orchestration for Iterative Visual Reasoning (ICLR 2026 · similarity 0.90)
- Through the Lens of Contrast: Self-Improving Visual Reasoning in VLMs (ICLR 2026 · similarity 0.90)
- MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
