Research map / Detection and segmentation / 3D detection and driving
SWAN: World-Aware Adaptive Multimodal Networks for Runtime Variations
Jason Wu, Shir-Kang Jin, Yuyang Yuan, Maggie Wigness, Lance Kaplan, Hang Qiu, Mani Srivastava
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 QoI-aware controller to Naive allocation (ablation) | SWAN · LiDAR Beamsreduce, 4-layer budget | mAP: 3.88 -> 16.12 | Improved |
| use NeuralSort controller instead of ADMN straight-through controller | SWAN-C · LiDAR Beamsreduce, 4-layer budget | mAP: 7.48 -> 16.12 | Improved |
| use NeuralSort controller instead of ADMN | SWAN-C · LiDAR Motionblur, 6-layer budget | NDS: 48.05 -> 56.19 | Improved |
| add SkipGate to the controller at 16-layer budget (ablation) | SWAN · 16-layer budget | GFLOPs: 556.41 -> 455.66 | Improved |
| add SkipGate to the controller at 16-layer budget (ablation) | SWAN · 16-layer budget, RTX 4090 | latency (ms): 120.93 -> 116.17 | No clear effect |
| add SkipGate to the controller at 4-layer budget (ablation) | SWAN · LiDAR Beamsreduce, 4-layer budget | mAP: 16.12 -> 16.11 | No clear effect |
| add SkipGate to the controller at 16-layer budget (ablation) | SWAN · LiDAR Beamsreduce, 16-layer budget | NDS / mAP: 40.97 -> 39.01 NDS, 27.74 -> 28.37 mAP | Mixed |
| add token pruning on top of SkipGate (ablation) | SWAN · 4-layer budget, RTX 4090 | latency (ms): 104.33 -> 72.21 | Improved |
| add token pruning on top of SkipGate (ablation) | SWAN · LiDAR Beamsreduce, 4-layer budget | mAP: 16.11 -> 14.18 | Hurt |
| add SkipGate on Jetson Orin under 30W power (ablation) | SWAN · Jetson Orin camera fog | latency reduction: 16.5% | Improved |
| run full SWAN with token pruning on Jetson Orin (ablation) | SWAN-PSC · Jetson Orin camera fog | latency reduction: over 50% | Improved |
| use 4-layer SWAN versus fully-provisioned base network | SWAN-PSC · Camera Motionblur | NDS / mAP: within 2.7 NDS and 5.4 mAP | Mixed |
Closest papers by meaning
- ADMN: A Layer-Wise Adaptive Multimodal Network for Dynamic Input Noise and Compute Resources (NeurIPS 2025 · similarity 0.87)
- M3Net: Multimodal Multi-task Learning for 3D Detection, Segmentation, and Occupancy Prediction in Autonomous Driving (AAAI 2025 · similarity 0.83)
- Towards Robust Multimodal Open-set Test-time Adaptation via Adaptive Entropy-aware Optimization (ICLR 2025 · similarity 0.83)
- Genesis: Multimodal Driving Scene Generation with Spatio-Temporal and Cross-Modal Consistency (NeurIPS 2025 · similarity 0.82)
- H-MBA: Hierarchical MamBa Adaptation for Multi-Modal Video Understanding in Autonomous Driving (AAAI 2025 · similarity 0.82)
- Lyapunov-Stable Adaptive Control for Multimodal Concept Drift (NeurIPS 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.
