Research map / Vision-language and multimodal / Cross-modal learning
Reducing Unimodal Bias in Multi-Modal Semantic Segmentation with Multi-Scale Functional Entropy Regularization
Xu Zheng, Yuanhuiyi Lyu, Lutao Jiang, Danda Pani Paudel, Luc Van Gool, Xuming Hu
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 prediction-level functional-entropy regularization R_p to supervised loss (ablation) | Ours (multi-modal segmentation) · MUSES | mean mIoU: 42.60 -> 50.02 | Improved |
| add prediction-level regularization R_p (ablation) | Ours (multi-modal segmentation) · MUSES (event only) | mIoU: 6.48 -> 29.44 | Improved |
| add prediction-level regularization R_p (ablation) | Ours (multi-modal segmentation) · MUSES (frame only) | mIoU: 60.63 -> 59.63 | Hurt |
| add multi-scale feature-level regularization R_f on top of R_p (ablation) | Ours (multi-modal segmentation) · MUSES | mean mIoU: 50.02 -> 50.44 | Improved |
| add both regularizers (R_p and R_f) (ablation) | Ours (multi-modal segmentation) · MUSES | mean accuracy: 52.53 -> 61.48 | Improved |
| vary lambda_p from 0.0 to 0.3 (ablation) | Ours (R_p weight) · MUSES | mean mIoU: 48.48 -> 49.89 | Improved |
| increase lambda_p from 0.3 to 0.7 (ablation) | Ours (R_p weight) · MUSES | mean mIoU: 49.89 -> 48.55 | Hurt |
| add feature regularization weight lambda_f = 0.02 (ablation) | Ours (R_f weight) · MUSES | mean mIoU: 49.89 -> 50.44 | Improved |
| increase lambda_f from 0.02 to 0.06 (ablation) | Ours (R_f weight) · MUSES | mean mIoU: 50.44 -> 50.03 | Hurt |
| use the original VQA functional-entropy regularization directly instead of the multi-scale one (ablation) | Ours (regularization design) · MUSES | mIoU: 50.44 -> 42.89 | Hurt |
| add regularization terms (training cost) (ablation) | Ours · Training on 8 H100 GPUs | training time (hours): 2.462 -> 2.476 | No clear effect |
| replace MAGIC with the proposed regularized model | MAGIC · DELIVER (four modalities) | mean mIoU: 40.49 -> 48.05 | 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.
- MAGIC — MAGIC: Multimodal relAtional Graph adversarIal inferenCe for Diverse and Unpaired Text-Based Image Captioning (AAAI 2022) · better on 6 of 7 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
- Keep the Balance: A Parameter-Efficient Symmetrical Framework for RGB+X Semantic Segmentation (CVPR 2025 · similarity 0.88)
- OmniSegmentor: A Flexible Multi-Modal Learning Framework for Semantic Segmentation (NeurIPS 2025 · similarity 0.88)
- UniMRSeg: Unified Modality-Relax Segmentation via Hierarchical Self-Supervised Compensation (NeurIPS 2025 · similarity 0.86)
- Unsupervised Modality Adaptation with Text-to-Image Diffusion Models for Semantic Segmentation (NeurIPS 2024 · similarity 0.85)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.85)
- Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning (NeurIPS 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.
