Research map / Detection and segmentation / Open-set and few-shot detection
Latent Expression Generation for Referring Image Segmentation and Grounding
Seonghoon Yu, Joonbeom Hong, Joonseok Lee, Jeany Son
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 latent expressions (naive) (ablation) | Latent-VG base · RefCOCO+ val | mIoU: 68.76 -> 71.03 | Improved |
| add subject distributor to latent expressions (ablation) | Latent-VG · RefCOCO+ val | mIoU: 71.03 -> 71.59 | Improved |
| add visual concept injector to latent expressions (ablation) | Latent-VG · RefCOCO+ val | mIoU: 71.03 -> 71.86 | Improved |
| combine subject distributor and visual concept injector (ablation) | Latent-VG · RefCOCO+ val | mIoU: 71.03 -> 72.63 | Improved |
| add positive-margin contrastive loss (ablation) | Latent-VG · RefCOCO+ val | mIoU: 72.63 -> 73.19 | Improved |
| replace InfoNCE with positive-margin contrastive loss (ablation) | Latent-VG · RefCOCO+ val | mIoU: 72.03 -> 73.19 | Improved |
| use margin 0.2 instead of no margin (ablation) | Latent-VG · RefCOCO+ val | mIoU: 72.03 -> 73.19 | Improved |
| add semantic dropout in latent expression initialization (ablation) | Latent-VG · RefCOCO+ val | mIoU: 72.33 -> 73.19 | Improved |
| use mean-similarity threshold instead of no threshold for target patches (ablation) | Latent-VG · RefCOCO+ val | mIoU: 72.03 -> 73.19 | Improved |
| use orthogonal initialization for concept tokens (ablation) | Latent-VG · RefCOCO+ val | mIoU: 72.59 -> 73.19 | Improved |
| increase latent expressions from 2 to 3 ({4,10} to {4,10,16}) (ablation) | Latent-VG · RefCOCO+ val | mIoU: 73.19 -> 72.45 | Hurt |
| use varied lengths {4,10} instead of equal lengths {4,4} (ablation) | Latent-VG · RefCOCO+ val | mIoU: 73.09 -> 73.19 | Improved |
Closest papers by meaning
- Locate Then Segment: A Strong Pipeline for Referring Image Segmentation (CVPR 2021 · similarity 0.85)
- Mask Grounding for Referring Image Segmentation (CVPR 2024 · similarity 0.85)
- Unveiling Parts Beyond Objects: Towards Finer-Granularity Referring Expression Segmentation (CVPR 2024 · similarity 0.85)
- Multimodal Referring Segmentation: A Survey (IJCV 2026 · similarity 0.84)
- Early Semantic Grounding in Image Editing Models for Zero-Shot Referring Image Segmentation (NeurIPS 2026 · similarity 0.84)
- Boosting Weakly Supervised Referring Image Segmentation via Progressive Comprehension (NeurIPS 2024 · similarity 0.83)
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.
