Research map / Vision-language and multimodal / Human and embodied modeling
Dynamic Inference With Grounding Based Vision and Language 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 |
|---|---|---|---|
| Apply learned dynamic layer skipping, token pruning, and token fusion | ViTMDETR · RefCOCO val | accuracy: 87.6 -> 87.4 | No clear effect |
| Apply learned dynamic layer skipping, token pruning, and token fusion | ViTMDETR · Image-language tasks | frames per second: 27 -> 41 (↑50%) | Helped |
| Apply learned dynamic layer skipping, token pruning, and token fusion | ViTMDETR · Image-language tasks | GFLOPs: 60 -> 32 (↓50%) | Helped |
| Use dynamic layer skipping only (ablation) | ViTMDETR · Image-language tasks | frames per second: 27 -> 34 | Helped |
| Use dynamic token pruning and fusion only (ablation) | ViTMDETR · Image-language tasks | frames per second: 27 -> 32 | Helped |
| Combine dynamic layer skipping with token pruning and fusion (ablation) | ViTMDETR · Image-language tasks | frames per second: 27 -> 41 | Helped |
| Add knowledge distillation from ViTMDETR (ablation) | D-ViTMDETR · RefCOCO val | accuracy: 86.9 -> 87.4 | Helped |
| Pre-train the decision network (ablation) | D-ViTMDETR · RefCOCO val | accuracy: 77.4 -> 84.9 | Helped |
| Use 64x64 token-clustering windows instead of 16x16 windows (ablation) | D-ViTMDETR · RefCOCO val | accuracy: 84.5 -> 87.4 | Helped |
| Use 128x128 token-clustering windows instead of 64x64 windows (ablation) | D-ViTMDETR · RefCOCO val | accuracy: 87.4 -> 87.0 | Hurt |
| Use 32x32 token-clustering windows instead of 64x64 windows (ablation) | D-ViTMDETR · RefCOCO val | accuracy: 87.4 -> 85.8 | Hurt |
Closest papers by meaning
- GeM-VG: Towards Generalized Multi-image Visual Grounding with Multimodal Large Language Models (AAAI 2026 · similarity 0.89)
- Connecting the Dots: Training-Free Visual Grounding via Agentic Reasoning (AAAI 2026 · similarity 0.88)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.88)
- SimVG: A Simple Framework for Visual Grounding with Decoupled Multi-modal Fusion (NeurIPS 2024 · similarity 0.88)
- Dynamic-LLaVA: Efficient Multimodal Large Language Models via Dynamic Vision-language Context Sparsification (ICLR 2025 · similarity 0.88)
- VISion On Request: Enhanced VLLM efficiency with sparse, dynamically selected, vision-language interactions (CVPR 2026 · similarity 0.88)
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.