Research map / Label-efficient and robust learning / Recognition and long-tail
NexusFlow: Unifying Disparate Tasks under Partial Supervision via Invertible Flow Networks
Fangzhou Lin, Yuping Wang, Yuliang Guo, Zixun Huang, Xinyu Huang, Haichong Zhang, Kazunori Yamada, Zhengzhong Tu, Liu Ren, Ziming Zhang
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 invertible coupling layers (ablation) | NexusFlow · nuScenes PS-MTL | AMOTA: 0.214 -> 0.329 | Improved |
| w/o inv vs baseline on tracking (ablation) | NexusFlow · nuScenes PS-MTL | AMOTA: 0.289 -> 0.214 | Hurt |
| increase coupling depth from 4 to 6 layers (ablation) | NexusFlow · nuScenes PS-MTL | AMOTA: 0.292 -> 0.329 | Improved |
| increase coupling depth from 6 to 8 layers (ablation) | NexusFlow · nuScenes PS-MTL | AMOTA: 0.329 -> 0.247 | Hurt |
| add NexusFlow alignment to UniAD (ablation) | UniAD · nuScenes PS-MTL | AMOTA: 0.289 -> 0.329 | Improved |
| add NexusFlow alignment to UniAD (ablation) | UniAD · nuScenes PS-MTL | Lanes IoU (%): 27.1 -> 37.1 | Improved |
| add NexusFlow alignment to UniAD (ablation) | UniAD · nuScenes PS-MTL | Drivable IoU (%): 62.7 -> 64.5 | Mixed |
| add prior PS-MTL methods (MTPSL, JTR) designed for homogeneous tasks (ablation) | UniAD · nuScenes PS-MTL | AMOTA: 0.289 -> 0.255 (MTPSL), 0.197 (JTR) | Hurt |
| two-stage fine-tuning instead of one-phase joint training (ablation) | NexusFlow · nuScenes PS-MTL | AMOTA: 0.318 -> 0.329 | Improved |
| center-based instead of pairwise alignment loss (ablation) | NexusFlow · NYUv2 | Segmentation IoU: 31.35 -> 31.70 | No clear effect |
| add NexusFlow to MTL on NYUv2 (ablation) | MTL baseline · NYUv2 | Segmentation IoU: 26.75 -> 31.70 | Improved |
| add NexusFlow vs MTPSL on memory | NexusFlow · NYUv2 | GPU memory: 8900 MiB -> 4554 MiB | Improved |
Closest papers by meaning
- Weakly Supervised Learning of Rigid 3D Scene Flow (CVPR 2021 · similarity 0.84)
- Just Go With the Flow: Self-Supervised Scene Flow Estimation (CVPR 2020 · similarity 0.84)
- M3Net: Multimodal Multi-task Learning for 3D Detection, Segmentation, and Occupancy Prediction in Autonomous Driving (AAAI 2025 · similarity 0.83)
- DualAD: Disentangling the Dynamic and Static World for End-to-End Driving (CVPR 2024 · similarity 0.83)
- Amalgamating Multi-Task Models with Heterogeneous Architectures (AAAI 2024 · similarity 0.82)
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 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.
