Research map / Probabilistic and scientific modeling / Dynamical systems
NeuroMamba: A Universal Spatiotemporal Module for Robust Perception in Degraded Sensory Streams
Jinfeng Li, Huijia Song, Xiangyue Hu, HanLiang Zhou, Jiahui Zhang, Xinpeng Jiang, Fangli Guan, Bin Lin, Dingran Dong, Liqi Yan, Pan Li
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 Mamba-based HSR module (ablation) | BEVFusion · nuScenes-D Medium | mAP: 35.4 -> 41.3 | Improved |
| add Spiking Confidence Gate on top of HSR (ablation) | BEVFusion + HSR · nuScenes-D Medium | mAP: 41.3 -> 52.5 | Improved |
| use spiking gate instead of Sigmoid MLP gate (ablation) | HSR · nuScenes-D Medium | mAP: 36.5 -> 52.5 | Improved |
| use spiking gate instead of Transformer/GRU/CNN gates of similar size (ablation) | HSR · nuScenes-D Medium | mAP: 44.2 (Transformer), 43.3 (GRU), 40.5 (CNN) -> 52.5 | Improved |
| firing threshold Vth set to 1.0 rather than 0.5 or 2.0 (ablation) | SCG · nuScenes-D Medium | mAP: 49.5 (0.5), 46.0 (2.0) -> 52.5 (1.0) | Improved |
| use single patch scale 6 instead of multi-scale (2, 4, 6) (ablation) | NeuroMamba · nuScenes-D Medium | mAP: 50.2 -> 52.5 | Improved |
| replace Transformer with Mamba in HSR (ablation) | HSR · nuScenes-D Medium, T=5 | mAP: 50.7 -> 52.5 | Improved |
| extend temporal window from T=5 to T=20 (ablation) | Mamba HSR · nuScenes-D Medium | mAP: 52.5 -> 47.0 | Hurt |
| add NeuroMamba plug-in to BEVFusion on clean data | BEVFusion · nuScenes-D Clean | NDS: 59.7 -> 58.2 | Hurt |
| use NeuroMamba plug-in instead of BEVDiffuser | BEVFusion · nuScenes-D Medium | mAP: 38.1 -> 52.5 | Improved |
| use NeuroMamba plug-in instead of BEVDiffuser (efficiency) | BEVFusion · inference cost | latency (ms): 84.19 -> 8.27 | Improved |
| add BEVDiffuser plug-in under heavy and extreme degradation | BEVFusion · nuScenes-D Extreme | mAP: 7.8 -> 5.8 | Hurt |
Closest papers by meaning
- Unveiling the Spatial-temporal Effective Receptive Fields of Spiking Neural Networks (NeurIPS 2025 · similarity 0.85)
- STSBench: A Large-Scale Dataset for Modeling Neuronal Activity in the Dorsal Stream of Primate Visual Cortex (NeurIPS 2025 · similarity 0.84)
- Spatio-Temporal Interactive Learning for Efficient Image Reconstruction of Spiking Cameras (NeurIPS 2024 · similarity 0.84)
- A Scalable, Causal, and Energy Efficient Framework for Neural Decoding with Spiking Neural Networks (NeurIPS 2025 · similarity 0.84)
- Neurons as Detectors of Coherent Sets in Sensory Dynamics (NeurIPS 2025 · similarity 0.83)
- SpikeNeRF: Learning Neural Radiance Fields from Continuous Spike Stream (CVPR 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.
