Research map / Vision-language and multimodal / Human and embodied modeling
What’s Left? Concept Grounding with Logic-Enhanced Foundation Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use LEFT instead of the strongest end-to-end baseline | CLEVR concept learning · CLEVR | accuracy: 0.989 -> 0.996 | Helped |
| Use LEFT instead of BUTD | 3D referring expression comprehension · SR3D | accuracy: 0.670 -> 0.620 | Hurt |
| Use LEFT instead of NSPose | motion question answering · HumanMotionQA | accuracy: 0.578 -> 0.582 | Helped |
| Use LEFT instead of ProgramPort | robotic sequential packing · CLIPort | average total reward: 0.797 -> 0.793 | Hurt |
| Use LEFT instead of MDETR | natural-language visual question answering · CLEVR-Humans | accuracy: 0.817 vs 0.788 / 0.698 | Hurt |
| Use LEFT instead of NS3D | natural-language 3D referring expression comprehension · NR3D | accuracy: 0.526 vs 0.496 / 0.485 | Hurt |
| Use LEFT instead of end-to-end methods with limited training data | SR3D concept learning · SR3D, 0.5% training data | accuracy: LEFT 0.410; NS3D 0.426; end-to-end methods 0.083, 0.161, 0.172 | Mixed |
| Use LEFT instead of NSCL | CLEVR with limited training data · CLEVR, 10% training data | accuracy: 0.989 -> 0.941 | Hurt |
| Generate programs with the LLM interpreter rather than use ground-truth programs (ablation) | LEFT zero-shot transfer · CLEVR-Ref | accuracy: 1.00 -> 0.94 | Hurt |
| Generate programs with the LLM interpreter rather than use ground-truth programs (ablation) | LEFT zero-shot transfer · CLEVR-Puzzles | accuracy: 0.92 -> 0.75 | Hurt |
| Generate programs with the LLM interpreter rather than use ground-truth programs (ablation) | LEFT zero-shot transfer · CLEVR-RPM | accuracy: 1.00 -> 0.87 | Hurt |
| Use LEFT with LLM-generated programs instead of VisProg | zero-shot visual reasoning · CLEVR-Ref | accuracy: 0.35 -> 0.94 | Helped |
Closest papers by meaning
- What Makes a Maze Look Like a Maze? (ICLR 2025 · similarity 0.88)
- Naturally Supervised 3D Visual Grounding with Language-Regularized Concept Learners (CVPR 2024 · similarity 0.88)
- No Labels, No Problem: Training Visual Reasoners with Multimodal Verifiers (ICLR 2026 · similarity 0.87)
- Connecting the Dots: Training-Free Visual Grounding via Agentic Reasoning (AAAI 2026 · similarity 0.87)
- MentisOculi: Revealing the Limits of Reasoning with Mental Imagery (ICML 2026 · similarity 0.87)
- Learning to See Before Seeing: Demystifying LLM Visual Priors from Language Pre-training (ICLR 2026 · similarity 0.87)
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.