Research map / Detection and segmentation / Open-set and few-shot detection
Referring Expression Comprehension for Small Objects
Kanoko Goto, Takumi Hirose, Mahiro Ukai, Shuhei Kurita, Nakamasa Inoue
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 PIZA module (zooming-step embedding) to Adapter+ (ablation) | Adapter+ · SOREC Val (Train-S) | mAcc: 26.0 -> 36.8 | Improved |
| add PIZA module to LoRA (ablation) | LoRA · SOREC Val (Train-S) | mAcc: 21.6 -> 30.9 | Improved |
| add PIZA module to CoOp (ablation) | CoOp · SOREC Val (Train-S) | mAcc: 20.2 -> 26.3 | Improved |
| add PIZA module to VPT (ablation) | VPT · SOREC Val (Train-S) | mAcc: 19.8 -> 26.5 | Improved |
| add zooming-step embedding insertion to PIZA-Adapter+ (ablation) | PIZA-Adapter+ · SOREC Val (Train-S) | mAcc: 36.7 -> 36.8 | No clear effect |
| add zooming-step embedding insertion to PIZA-LoRA (ablation) | PIZA-LoRA · SOREC Val (Train-S) | mAcc: 30.2 -> 30.9 | Improved |
| predict an adaptive number of zooming steps (avg 2.11) instead of forcing T=1 (ablation) | PIZA-Adapter+ · SOREC Val (Train-S) | mAcc: 25.8 -> 36.8 | Improved |
| adaptive zooming steps instead of forcing T=3 (ablation) | PIZA-Adapter+ · SOREC Val (Train-S) | mAcc: 34.3 -> 36.8 | Improved |
| increase adapter bottleneck dimension from 32 to 256 (ablation) | PIZA-Adapter+ · SOREC Val (Train-S) | mAcc: 35.1 -> 36.8 | Improved |
| use pre-trained weights instead of training from scratch (ablation) | Full fine-tuning · SOREC Val (Train-S) | mAcc: 0.00 -> 29.5 | Improved |
| train on Train-L instead of Train-S | PIZA-Adapter+ · SOREC Val | mAcc: 36.8 -> 39.0 | Improved |
| increase LoRA rank from 4 to 64 (ablation) | PIZA-LoRA · SOREC Val (Train-S) | mAcc: 30.8 -> 31.1 | No clear effect |
Closest papers by meaning
- ScanFormer: Referring Expression Comprehension by Iteratively Scanning (CVPR 2024 · similarity 0.89)
- Rethinking Two-Stage Referring Expression Comprehension: A Novel Grounding and Segmentation Method Modulated by Point (AAAI 2024 · similarity 0.89)
- RefDetector: A Simple Yet Effective Matching-based Method for Referring Expression Comprehension (AAAI 2025 · similarity 0.88)
- DiffusionREC: Diffusion Model with Adaptive Condition for Referring Expression Comprehension (AAAI 2025 · similarity 0.88)
- Boosting Weakly Supervised Referring Image Segmentation via Progressive Comprehension (NeurIPS 2024 · similarity 0.88)
- A Large-Scale Human-Centric Benchmark for Referring Expression Comprehension in the LMM Era (NeurIPS 2024 · 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.
