Research map / Architectures and efficiency / Attention and transformers
An Efficient and Effective Transformer Decoder-Based Framework for Multi-Task Visual Grounding
Wei Chen, Long Chen, Yu Wu
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 |
|---|---|---|---|
| use transformer decoder instead of encoder for vision-language fusion (ablation) | EEVG · RefCOCOg val(U) REC | accuracy (%): 78.43 -> 79.27 | Improved |
| use decoder instead of encoder for fusion (ablation) | EEVG · runtime, 196 visual tokens, 300 linguistic tokens | runtime (ms): 14.85 -> 8.09 | Improved |
| static token elimination instead of no elimination (ablation) | EEVG · RefCOCOg RES val | mIoU: 68.16 -> 66.38 | Hurt |
| dynamic token elimination instead of no elimination (ablation) | EEVG · RefCOCOg RES val | mIoU: 68.16 -> 68.89 | Improved |
| add adaptive spatial attention to dynamic elimination (ablation) | EEVG · RefCOCOg RES val | mIoU: 68.89 -> 69.15 | Improved |
| use efficient light mask head instead of FPN-like (ablation) | EEVG · RefCOCOg | parameters and RES val mIoU: 8.11M -> 0.79M params; 68.64 -> 69.15 mIoU | Improved |
| add 1-channel convolution layer to mask head (ablation) | EEVG mask head · RefCOCOg RES val | mIoU: 67.71 -> 69.15 | Improved |
| EEVG versus prior methods on RefCOCOg (complex expressions) | EEVG ViT-B · RefCOCOg | accuracy (%): +3.62% val, +3.93% test | Improved |
| EEVG with pre-training on large corpus versus PolyFormer on RefCOCO+ REC | EEVG · RefCOCO+ val REC (pre-trained) | accuracy (%): 83.73 vs 82.24 | Hurt |
Closest papers by meaning
- SimVG: A Simple Framework for Visual Grounding with Decoupled Multi-modal Fusion (NeurIPS 2024 · similarity 0.92)
- VGent: Visual Grounding via Modular Design for Disentangling Reasoning and Prediction (CVPR 2026 · similarity 0.89)
- GeM-VG: Towards Generalized Multi-image Visual Grounding with Multimodal Large Language Models (AAAI 2026 · similarity 0.88)
- Referencing Where to Focus: Improving Visual Grounding with Referential Query (NeurIPS 2024 · similarity 0.88)
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.88)
- Look Around Before Locating: Considering Content and Structure Information for Visual Grounding (AAAI 2025 · 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.
