Research map / Vision-language and multimodal / Multimodal LLMs
Learning to Compose and Reason with Language Tree Structures for Visual Grounding
Richang Hong, Daqing Liu, Xiaoyu Mo, Xiangnan He, Hanwang Zhang
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 |
|---|---|---|---|
| replace holistic BiLSTM chain encoder with parser tree (TreeLSTM) (ablation) | Chain · RefCOCOg test (GT boxes) | Top-1 accuracy (%): 71.69 -> 73.07 | Improved |
| learn tree with supervised pre-training instead of from scratch (ablation) | RvG-Tree-Scratch · RefCOCOg test (GT boxes) | Top-1 accuracy (%): 72.00 -> 75.20 | Improved |
| add in-node score (ablation) | RvG-Tree/Node · RefCOCOg test (GT boxes) | Top-1 accuracy (%): 74.36 -> 75.20 | Improved |
| add recursive score accumulation from score node (ablation) | RvG-Tree/S · RefCOCOg test (GT boxes) | Top-1 accuracy (%): 73.18 -> 75.20 | Improved |
| add pairwise score from feature node (ablation) | RvG-Tree/F · RefCOCOg test (GT boxes) | Top-1 accuracy (%): 73.98 -> 75.20 | Improved |
| add recursive tree-structured grounding score on top of tree encoder (ablation) | RvG-Tree-Fix · RefCOCOg val (GT boxes) | Top-1 accuracy (%): 73.69 -> 76.82 | Improved |
| add pairwise score from feature node (ablation) | RvG-Tree/F · RefCOCOg val (GT boxes) | Top-1 accuracy (%): 74.12 -> 76.82 | Improved |
| add in-node score (ablation) | RvG-Tree/Node · RefCOCO+ testB (detected boxes) | Top-1 accuracy (%): 55.72 -> 56.66 | Improved |
| learn tree from scratch instead of expert pretraining (ablation) | RvG-Tree-Scratch · RefCOCO val (GT boxes) | Top-1 accuracy (%): 83.48 -> 79.65 | Hurt |
| replace MMI cross-entropy loss with triplet loss with hard negatives (ablation) | RvG-Tree · RefCOCO family | Top-1 accuracy (%) | No clear effect |
| use random instead of GloVe word initialization (ablation) | RvG-Tree · RefCOCO family | Top-1 accuracy (%) | No clear effect |
| use REINFORCE instead of straight-through Gumbel-Softmax (ablation) | RvG-Tree · RefCOCO family | Top-1 accuracy (%): 1e-6 | Hurt |
Closest papers by meaning
- Connecting the Dots: Training-Free Visual Grounding via Agentic Reasoning (AAAI 2026 · similarity 0.90)
- Investigating Compositional Challenges in Vision-Language Models for Visual Grounding (CVPR 2024 · similarity 0.89)
- Look Around Before Locating: Considering Content and Structure Information for Visual Grounding (AAAI 2025 · similarity 0.89)
- GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation (CVPR 2026 · similarity 0.89)
- SimVG: A Simple Framework for Visual Grounding with Decoupled Multi-modal Fusion (NeurIPS 2024 · similarity 0.89)
- GeM-VG: Towards Generalized Multi-image Visual Grounding with Multimodal Large Language Models (AAAI 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.
