Research map / Vision-language and multimodal / Multimodal LLMs
Different Changes Require Different Reasoning: Change-Type-Specialized Experts for Robust Change Captioning
Jiyoung Park, InJae Oh, Jung Uk Kim
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 MEDIC change-type experts | DIRL · CLEVR-DC | CIDEr: 84.1 -> 99.8 | Improved |
| add MEDIC | SCORER · Spot-the-Diff (single-change) | CIDEr: 38.9 -> 39.2 | No clear effect |
| add MEDIC | SMART · CLEVR-Change | CIDEr: 127.0 -> 128.1 | Improved |
| add MEDIC in the multi-change setting | DIRL · Spot-the-Diff (multi-change) | CIDEr: 21.5 -> 31.2 | Improved |
| use memory expert instead of 2-layer MLP expert (ablation) | DIRL+MEDIC · CLEVR-DC | CIDEr: 96.1 -> 99.8 | Improved |
| add expert disentangle loss (ablation) | DIRL+MEDIC · CLEVR-DC | CIDEr: 98.2 -> 99.8 | Improved |
| add expert consistency loss (ablation) | DIRL+MEDIC · CLEVR-DC | CIDEr: 99.1 -> 99.8 | Improved |
| add routing loss (ablation) | DIRL+MEDIC · CLEVR-DC | CIDEr: 87.7 -> 96.7 | Improved |
| increase memory slots from 100 to 400 (ablation) | DIRL+MEDIC · CLEVR-DC | CIDEr: 99.8 -> 97.7 | Hurt |
| use type-specific experts rather than auxiliary type classification loss (ablation) | DIRL · CLEVR-DC | CIDEr: 87.2 -> 99.8 | Improved |
| use two-stage routing instead of single-stage (ablation) | DIRL+MEDIC · CLEVR-DC | CIDEr: 90.4 -> 99.8 | Improved |
| corrupt 40% of training type labels (ablation) | DIRL+MEDIC · CLEVR-DC | CIDEr: 99.8 -> 94.0 | Hurt |
Closest papers by meaning
- VisMin: Visual Minimal-Change Understanding (NeurIPS 2024 · similarity 0.88)
- Change3D: Revisiting Change Detection and Captioning from A Video Modeling Perspective (CVPR 2025 · similarity 0.87)
- Region-aware Difference Distilling with Attribute-guided Contrastive Regularization for Change Captioning (AAAI 2025 · similarity 0.87)
- Describing Differences in Image Sets with Natural Language (CVPR 2024 · similarity 0.86)
- VIXEN: Visual Text Comparison Network for Image Difference Captioning (AAAI 2024 · similarity 0.86)
- DECIDER: Difference-aware Contrastive Diffusion Model with Adversarial Perturbations for Image Change Captioning (AAAI 2025 · similarity 0.85)
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.
