Research map / Detection and segmentation / CNN features
WaveMamba: Wavelet-Driven Mamba Fusion for RGB-Infrared Object Detection
Haodong Zhu, Wenhao Dong, Linlin Yang, Hong Li, Yuguang Yang, Yangyang Ren, Qingcheng Zhu, Zichao Feng, Changbai Li, Shaohui Lin, Runqi Wang, Xiaoyan Luo and 1 more
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 |
|---|---|---|---|
| replace original YOLOv8 head with IDWT-based improved head (ablation) | WaveMamba · M3FD | mAP50: 91.0 -> 92.1 | Improved |
| add Shallow Fusion Mamba (channel swap + VSS) module (ablation) | WaveMamba · M3FD | mAP50: 90.6 -> 92.1 | Improved |
| add Deep Fusion Mamba (gated attention) module (ablation) | WaveMamba · M3FD | mAP50: 90.2 -> 92.1 | Improved |
| fuse in wavelet domain with average fusion instead of plain average fusion without DWT (ablation) | RGB-IR fusion baseline · M3FD | mAP50: 83.2 -> 86.6 | Improved |
| use LMFB for low-frequency fusion instead of average (ablation) | WaveMamba · M3FD | mAP50: 86.6 -> 90.6 | Improved |
| use HFE absolute-maximum fusion for high-frequency instead of average (ablation) | WaveMamba · M3FD | mAP50: 90.6 -> 91.0 | Improved |
| use three WMFBs instead of one (ablation) | WaveMamba · M3FD | mAP50: 91.3 -> 92.1 | Improved |
| place WMFBs at stages P2, P3, P5 rather than other combinations (ablation) | WaveMamba · M3FD | mAP50: 91.8 -> 92.1 | Improved |
| use Haar wavelet instead of other wavelet bases (MEYR, SYM3, COIF3, DB3) (ablation) | WaveMamba · M3FD | mAP50: 91.7-91.8 -> 92.1 | No clear effect |
| fuse RGB and IR with WaveMamba instead of single-modality YOLOv8l | YOLOv8 · M3FD | mAP50: 80.9 -> 92.1 | Improved |
| WaveMamba vs CrossFormer with YOLOv5 backbone on FLIR-Aligned (efficiency) | WaveMamba · FLIR-Aligned | inference time (ms): 80.0 -> 44.1 | Improved |
Later papers that compared with this method
2 later accepted papers report WaveMamba as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- InfraNet: Quality-Aware RGB Guidance for Infrared Object Detection (ECCV 2026 · better on 7 of 8 shared results)
- Fully Rotation-Equivariant Spectral-Spatial Learning for Multispectral Object Detection (ECCV 2026 · 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
- Efficient RGB-T Object Detection via Sparse Cross-Modality Fusion (ECCV 2026 · similarity 0.86)
- M3SR: Multi-Scale Multi-Perceptual Mamba for Efficient Spectral Reconstruction (AAAI 2026 · similarity 0.85)
- Tri-Modal Fusion Transformers for UAV-based Object Detection (CVPR 2026 · similarity 0.85)
- IRMamba: Pixel Difference Mamba with Layer Restoration for Infrared Small Target Detection (AAAI 2025 · similarity 0.84)
- IRPruneDet: Efficient Infrared Small Target Detection via Wavelet Structure-Regularized Soft Channel Pruning (AAAI 2024 · similarity 0.83)
- Mamba YOLO: A Simple Baseline for Object Detection with State Space Model (AAAI 2025 · similarity 0.82)
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.
