Research map / Vision-language and multimodal / Multimodal LLMs
New Dataset and Methods for Fine-Grained Compositional Referring Expression Comprehension via Specialist-MLLM Collaboration
Xuzheng Yang, Junzhuo Liu, Peng Wang, Guoqing Wang, Yang Yang, Heng Tao Shen
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 Slow-Fast Adaptation routing between specialist and MLLM | CogVLM · FineCops-Ref | Precision@1 (All): 69.46 -> 73.57 | Improved |
| add focus-enhancement strategy to SFA | CogVLM(SFA) · FineCops-Ref | Precision@1 (All): 73.57 -> 75.03 | Improved |
| add SFA routing | Qwen2-VL-7B · Ref-Adv | Precision@1 (All): 79.83 -> 79.62 | No clear effect |
| add Candidate Region Selection (CRS) tuning | InternVL2.5-8B · FineCops-Ref | Precision@1 (All): 66.05 -> 78.58 | Improved |
| add CRS tuning | Qwen2-VL-7B · Ref-Reasoning | Precision@1 (All): 35.77 -> 40.45 | Improved |
| add focus-enhancement strategy | MDETR · Ref-Adv | Precision@1: 73.73 -> 72.12 | Hurt |
| add focus-enhancement strategy | InternVL2.5-8B · Ref-Adv | Precision@1: 75.92 -> 80.29 | Improved |
| add focus-enhancement strategy | MM-GDINO-L · Ref-Adv | Precision@1: 74.95 -> 74.55 | Hurt |
| use Grounding DINO instead of YOLO-World as difficulty router (ablation) | SFA · FineCops-Ref | Precision@1 (All): 71.53 -> 74.06 | Improved |
| use Grounding DINO instead of OV-DINO as difficulty router (ablation) | SFA · FineCops-Ref | Precision@1 (All): 70.39 -> 74.06 | Improved |
| train CRS on FineCops-Ref data instead of RefCOCO/+/g (ablation) | CRS Qwen2-VL-7B · FineCops-Ref | Precision@1 (All): 78.33 -> 81.38 | Improved |
| increase CRS instruction tuning data from 2k to 12k (ablation) | CRS Qwen2-VL-7B · FineCops-Ref | Precision@1 (All): 74.95 -> 78.84 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CogVLM — CogVLM: Visual Expert for Pretrained Language Models (NeurIPS 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks (ICLR 2026 · similarity 0.92)
- LLM-wrapper: Black-Box Semantic-Aware Adaptation of Vision-Language Models for Referring Expression Comprehension (ICLR 2025 · similarity 0.91)
- Rethinking Two-Stage Referring Expression Comprehension: A Novel Grounding and Segmentation Method Modulated by Point (AAAI 2024 · similarity 0.91)
- Revisiting Counterfactual Problems in Referring Expression Comprehension (CVPR 2024 · similarity 0.91)
- A Large-Scale Human-Centric Benchmark for Referring Expression Comprehension in the LMM Era (NeurIPS 2024 · similarity 0.90)
- ScanFormer: Referring Expression Comprehension by Iteratively Scanning (CVPR 2024 · similarity 0.90)
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.
